Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/proc/base.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * proc base directory handling functions
8 *
9 * 1999, Al Viro. Rewritten. Now it covers the whole per-process part.
10 * Instead of using magical inumbers to determine the kind of object
11 * we allocate and fill in-core inodes upon lookup. They don't even
12 * go into icache. We cache the reference to task_struct upon lookup too.
13 * Eventually it should become a filesystem in its own. We don't use the
14 * rest of procfs anymore.
15 *
16 *
17 * Changelog:
18 * 17-Jan-2005
19 * Allan Bezerra
20 * Bruna Moreira <bruna.moreira@indt.org.br>
21 * Edjard Mota <edjard.mota@indt.org.br>
22 * Ilias Biris <ilias.biris@indt.org.br>
23 * Mauricio Lin <mauricio.lin@indt.org.br>
24 *
25 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
26 *
27 * A new process specific entry (smaps) included in /proc. It shows the
28 * size of rss for each memory area. The maps entry lacks information
29 * about physical memory size (rss) for each mapped file, i.e.,
30 * rss information for executables and library files.
31 * This additional information is useful for any tools that need to know
32 * about physical memory consumption for a process specific library.
33 *
34 * Changelog:
35 * 21-Feb-2005
36 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
37 * Pud inclusion in the page table walking.
38 *
39 * ChangeLog:
40 * 10-Mar-2005
41 * 10LE Instituto Nokia de Tecnologia - INdT:
42 * A better way to walks through the page table as suggested by Hugh Dickins.
43 *
44 * Simo Piiroinen <simo.piiroinen@nokia.com>:
45 * Smaps information related to shared, private, clean and dirty pages.
46 *
47 * Paul Mundt <paul.mundt@nokia.com>:
48 * Overall revision about smaps.
49 */
50
51#include <linux/uaccess.h>
52
53#include <linux/errno.h>
54#include <linux/time.h>
55#include <linux/proc_fs.h>
56#include <linux/stat.h>
57#include <linux/task_io_accounting_ops.h>
58#include <linux/init.h>
59#include <linux/capability.h>
60#include <linux/file.h>
61#include <linux/fdtable.h>
62#include <linux/generic-radix-tree.h>
63#include <linux/string.h>
64#include <linux/seq_file.h>
65#include <linux/namei.h>
66#include <linux/mnt_namespace.h>
67#include <linux/mm.h>
68#include <linux/swap.h>
69#include <linux/rcupdate.h>
70#include <linux/kallsyms.h>
71#include <linux/stacktrace.h>
72#include <linux/resource.h>
73#include <linux/module.h>
74#include <linux/mount.h>
75#include <linux/security.h>
76#include <linux/ptrace.h>
77#include <linux/tracehook.h>
78#include <linux/printk.h>
79#include <linux/cache.h>
80#include <linux/cgroup.h>
81#include <linux/cpuset.h>
82#include <linux/audit.h>
83#include <linux/poll.h>
84#include <linux/nsproxy.h>
85#include <linux/oom.h>
86#include <linux/elf.h>
87#include <linux/pid_namespace.h>
88#include <linux/user_namespace.h>
89#include <linux/fs_struct.h>
90#include <linux/slab.h>
91#include <linux/sched/autogroup.h>
92#include <linux/sched/mm.h>
93#include <linux/sched/coredump.h>
94#include <linux/sched/debug.h>
95#include <linux/sched/stat.h>
96#include <linux/posix-timers.h>
97#include <linux/time_namespace.h>
98#include <linux/resctrl.h>
99#include <trace/events/oom.h>
100#include "internal.h"
101#include "fd.h"
102
103#include "../../lib/kstrtox.h"
104
105/* NOTE:
106 * Implementing inode permission operations in /proc is almost
107 * certainly an error. Permission checks need to happen during
108 * each system call not at open time. The reason is that most of
109 * what we wish to check for permissions in /proc varies at runtime.
110 *
111 * The classic example of a problem is opening file descriptors
112 * in /proc for a task before it execs a suid executable.
113 */
114
115static u8 nlink_tid __ro_after_init;
116static u8 nlink_tgid __ro_after_init;
117
118struct pid_entry {
119 const char *name;
120 unsigned int len;
121 umode_t mode;
122 const struct inode_operations *iop;
123 const struct file_operations *fop;
124 union proc_op op;
125};
126
127#define NOD(NAME, MODE, IOP, FOP, OP) { \
128 .name = (NAME), \
129 .len = sizeof(NAME) - 1, \
130 .mode = MODE, \
131 .iop = IOP, \
132 .fop = FOP, \
133 .op = OP, \
134}
135
136#define DIR(NAME, MODE, iops, fops) \
137 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
138#define LNK(NAME, get_link) \
139 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
140 &proc_pid_link_inode_operations, NULL, \
141 { .proc_get_link = get_link } )
142#define REG(NAME, MODE, fops) \
143 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
144#define ONE(NAME, MODE, show) \
145 NOD(NAME, (S_IFREG|(MODE)), \
146 NULL, &proc_single_file_operations, \
147 { .proc_show = show } )
148#define ATTR(LSM, NAME, MODE) \
149 NOD(NAME, (S_IFREG|(MODE)), \
150 NULL, &proc_pid_attr_operations, \
151 { .lsm = LSM })
152
153/*
154 * Count the number of hardlinks for the pid_entry table, excluding the .
155 * and .. links.
156 */
157static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
158 unsigned int n)
159{
160 unsigned int i;
161 unsigned int count;
162
163 count = 2;
164 for (i = 0; i < n; ++i) {
165 if (S_ISDIR(entries[i].mode))
166 ++count;
167 }
168
169 return count;
170}
171
172static int get_task_root(struct task_struct *task, struct path *root)
173{
174 int result = -ENOENT;
175
176 task_lock(task);
177 if (task->fs) {
178 get_fs_root(task->fs, root);
179 result = 0;
180 }
181 task_unlock(task);
182 return result;
183}
184
185static int proc_cwd_link(struct dentry *dentry, struct path *path)
186{
187 struct task_struct *task = get_proc_task(d_inode(dentry));
188 int result = -ENOENT;
189
190 if (task) {
191 task_lock(task);
192 if (task->fs) {
193 get_fs_pwd(task->fs, path);
194 result = 0;
195 }
196 task_unlock(task);
197 put_task_struct(task);
198 }
199 return result;
200}
201
202static int proc_root_link(struct dentry *dentry, struct path *path)
203{
204 struct task_struct *task = get_proc_task(d_inode(dentry));
205 int result = -ENOENT;
206
207 if (task) {
208 result = get_task_root(task, path);
209 put_task_struct(task);
210 }
211 return result;
212}
213
214/*
215 * If the user used setproctitle(), we just get the string from
216 * user space at arg_start, and limit it to a maximum of one page.
217 */
218static ssize_t get_mm_proctitle(struct mm_struct *mm, char __user *buf,
219 size_t count, unsigned long pos,
220 unsigned long arg_start)
221{
222 char *page;
223 int ret, got;
224
225 if (pos >= PAGE_SIZE)
226 return 0;
227
228 page = (char *)__get_free_page(GFP_KERNEL);
229 if (!page)
230 return -ENOMEM;
231
232 ret = 0;
233 got = access_remote_vm(mm, arg_start, page, PAGE_SIZE, FOLL_ANON);
234 if (got > 0) {
235 int len = strnlen(page, got);
236
237 /* Include the NUL character if it was found */
238 if (len < got)
239 len++;
240
241 if (len > pos) {
242 len -= pos;
243 if (len > count)
244 len = count;
245 len -= copy_to_user(buf, page+pos, len);
246 if (!len)
247 len = -EFAULT;
248 ret = len;
249 }
250 }
251 free_page((unsigned long)page);
252 return ret;
253}
254
255static ssize_t get_mm_cmdline(struct mm_struct *mm, char __user *buf,
256 size_t count, loff_t *ppos)
257{
258 unsigned long arg_start, arg_end, env_start, env_end;
259 unsigned long pos, len;
260 char *page, c;
261
262 /* Check if process spawned far enough to have cmdline. */
263 if (!mm->env_end)
264 return 0;
265
266 spin_lock(&mm->arg_lock);
267 arg_start = mm->arg_start;
268 arg_end = mm->arg_end;
269 env_start = mm->env_start;
270 env_end = mm->env_end;
271 spin_unlock(&mm->arg_lock);
272
273 if (arg_start >= arg_end)
274 return 0;
275
276 /*
277 * We allow setproctitle() to overwrite the argument
278 * strings, and overflow past the original end. But
279 * only when it overflows into the environment area.
280 */
281 if (env_start != arg_end || env_end < env_start)
282 env_start = env_end = arg_end;
283 len = env_end - arg_start;
284
285 /* We're not going to care if "*ppos" has high bits set */
286 pos = *ppos;
287 if (pos >= len)
288 return 0;
289 if (count > len - pos)
290 count = len - pos;
291 if (!count)
292 return 0;
293
294 /*
295 * Magical special case: if the argv[] end byte is not
296 * zero, the user has overwritten it with setproctitle(3).
297 *
298 * Possible future enhancement: do this only once when
299 * pos is 0, and set a flag in the 'struct file'.
300 */
301 if (access_remote_vm(mm, arg_end-1, &c, 1, FOLL_ANON) == 1 && c)
302 return get_mm_proctitle(mm, buf, count, pos, arg_start);
303
304 /*
305 * For the non-setproctitle() case we limit things strictly
306 * to the [arg_start, arg_end[ range.
307 */
308 pos += arg_start;
309 if (pos < arg_start || pos >= arg_end)
310 return 0;
311 if (count > arg_end - pos)
312 count = arg_end - pos;
313
314 page = (char *)__get_free_page(GFP_KERNEL);
315 if (!page)
316 return -ENOMEM;
317
318 len = 0;
319 while (count) {
320 int got;
321 size_t size = min_t(size_t, PAGE_SIZE, count);
322
323 got = access_remote_vm(mm, pos, page, size, FOLL_ANON);
324 if (got <= 0)
325 break;
326 got -= copy_to_user(buf, page, got);
327 if (unlikely(!got)) {
328 if (!len)
329 len = -EFAULT;
330 break;
331 }
332 pos += got;
333 buf += got;
334 len += got;
335 count -= got;
336 }
337
338 free_page((unsigned long)page);
339 return len;
340}
341
342static ssize_t get_task_cmdline(struct task_struct *tsk, char __user *buf,
343 size_t count, loff_t *pos)
344{
345 struct mm_struct *mm;
346 ssize_t ret;
347
348 mm = get_task_mm(tsk);
349 if (!mm)
350 return 0;
351
352 ret = get_mm_cmdline(mm, buf, count, pos);
353 mmput(mm);
354 return ret;
355}
356
357static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
358 size_t count, loff_t *pos)
359{
360 struct task_struct *tsk;
361 ssize_t ret;
362
363 BUG_ON(*pos < 0);
364
365 tsk = get_proc_task(file_inode(file));
366 if (!tsk)
367 return -ESRCH;
368 ret = get_task_cmdline(tsk, buf, count, pos);
369 put_task_struct(tsk);
370 if (ret > 0)
371 *pos += ret;
372 return ret;
373}
374
375static const struct file_operations proc_pid_cmdline_ops = {
376 .read = proc_pid_cmdline_read,
377 .llseek = generic_file_llseek,
378};
379
380#ifdef CONFIG_KALLSYMS
381/*
382 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
383 * Returns the resolved symbol. If that fails, simply return the address.
384 */
385static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
386 struct pid *pid, struct task_struct *task)
387{
388 unsigned long wchan;
389 char symname[KSYM_NAME_LEN];
390
391 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
392 goto print0;
393
394 wchan = get_wchan(task);
395 if (wchan && !lookup_symbol_name(wchan, symname)) {
396 seq_puts(m, symname);
397 return 0;
398 }
399
400print0:
401 seq_putc(m, '0');
402 return 0;
403}
404#endif /* CONFIG_KALLSYMS */
405
406static int lock_trace(struct task_struct *task)
407{
408 int err = mutex_lock_killable(&task->signal->exec_update_mutex);
409 if (err)
410 return err;
411 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
412 mutex_unlock(&task->signal->exec_update_mutex);
413 return -EPERM;
414 }
415 return 0;
416}
417
418static void unlock_trace(struct task_struct *task)
419{
420 mutex_unlock(&task->signal->exec_update_mutex);
421}
422
423#ifdef CONFIG_STACKTRACE
424
425#define MAX_STACK_TRACE_DEPTH 64
426
427static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
428 struct pid *pid, struct task_struct *task)
429{
430 unsigned long *entries;
431 int err;
432
433 /*
434 * The ability to racily run the kernel stack unwinder on a running task
435 * and then observe the unwinder output is scary; while it is useful for
436 * debugging kernel issues, it can also allow an attacker to leak kernel
437 * stack contents.
438 * Doing this in a manner that is at least safe from races would require
439 * some work to ensure that the remote task can not be scheduled; and
440 * even then, this would still expose the unwinder as local attack
441 * surface.
442 * Therefore, this interface is restricted to root.
443 */
444 if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
445 return -EACCES;
446
447 entries = kmalloc_array(MAX_STACK_TRACE_DEPTH, sizeof(*entries),
448 GFP_KERNEL);
449 if (!entries)
450 return -ENOMEM;
451
452 err = lock_trace(task);
453 if (!err) {
454 unsigned int i, nr_entries;
455
456 nr_entries = stack_trace_save_tsk(task, entries,
457 MAX_STACK_TRACE_DEPTH, 0);
458
459 for (i = 0; i < nr_entries; i++) {
460 seq_printf(m, "[<0>] %pB\n", (void *)entries[i]);
461 }
462
463 unlock_trace(task);
464 }
465 kfree(entries);
466
467 return err;
468}
469#endif
470
471#ifdef CONFIG_SCHED_INFO
472/*
473 * Provides /proc/PID/schedstat
474 */
475static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
476 struct pid *pid, struct task_struct *task)
477{
478 if (unlikely(!sched_info_on()))
479 seq_puts(m, "0 0 0\n");
480 else
481 seq_printf(m, "%llu %llu %lu\n",
482 (unsigned long long)task->se.sum_exec_runtime,
483 (unsigned long long)task->sched_info.run_delay,
484 task->sched_info.pcount);
485
486 return 0;
487}
488#endif
489
490#ifdef CONFIG_LATENCYTOP
491static int lstats_show_proc(struct seq_file *m, void *v)
492{
493 int i;
494 struct inode *inode = m->private;
495 struct task_struct *task = get_proc_task(inode);
496
497 if (!task)
498 return -ESRCH;
499 seq_puts(m, "Latency Top version : v0.1\n");
500 for (i = 0; i < LT_SAVECOUNT; i++) {
501 struct latency_record *lr = &task->latency_record[i];
502 if (lr->backtrace[0]) {
503 int q;
504 seq_printf(m, "%i %li %li",
505 lr->count, lr->time, lr->max);
506 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
507 unsigned long bt = lr->backtrace[q];
508
509 if (!bt)
510 break;
511 seq_printf(m, " %ps", (void *)bt);
512 }
513 seq_putc(m, '\n');
514 }
515
516 }
517 put_task_struct(task);
518 return 0;
519}
520
521static int lstats_open(struct inode *inode, struct file *file)
522{
523 return single_open(file, lstats_show_proc, inode);
524}
525
526static ssize_t lstats_write(struct file *file, const char __user *buf,
527 size_t count, loff_t *offs)
528{
529 struct task_struct *task = get_proc_task(file_inode(file));
530
531 if (!task)
532 return -ESRCH;
533 clear_tsk_latency_tracing(task);
534 put_task_struct(task);
535
536 return count;
537}
538
539static const struct file_operations proc_lstats_operations = {
540 .open = lstats_open,
541 .read = seq_read,
542 .write = lstats_write,
543 .llseek = seq_lseek,
544 .release = single_release,
545};
546
547#endif
548
549static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
550 struct pid *pid, struct task_struct *task)
551{
552 unsigned long totalpages = totalram_pages() + total_swap_pages;
553 unsigned long points = 0;
554 long badness;
555
556 badness = oom_badness(task, totalpages);
557 /*
558 * Special case OOM_SCORE_ADJ_MIN for all others scale the
559 * badness value into [0, 2000] range which we have been
560 * exporting for a long time so userspace might depend on it.
561 */
562 if (badness != LONG_MIN)
563 points = (1000 + badness * 1000 / (long)totalpages) * 2 / 3;
564
565 seq_printf(m, "%lu\n", points);
566
567 return 0;
568}
569
570struct limit_names {
571 const char *name;
572 const char *unit;
573};
574
575static const struct limit_names lnames[RLIM_NLIMITS] = {
576 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
577 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
578 [RLIMIT_DATA] = {"Max data size", "bytes"},
579 [RLIMIT_STACK] = {"Max stack size", "bytes"},
580 [RLIMIT_CORE] = {"Max core file size", "bytes"},
581 [RLIMIT_RSS] = {"Max resident set", "bytes"},
582 [RLIMIT_NPROC] = {"Max processes", "processes"},
583 [RLIMIT_NOFILE] = {"Max open files", "files"},
584 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
585 [RLIMIT_AS] = {"Max address space", "bytes"},
586 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
587 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
588 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
589 [RLIMIT_NICE] = {"Max nice priority", NULL},
590 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
591 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
592};
593
594/* Display limits for a process */
595static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
596 struct pid *pid, struct task_struct *task)
597{
598 unsigned int i;
599 unsigned long flags;
600
601 struct rlimit rlim[RLIM_NLIMITS];
602
603 if (!lock_task_sighand(task, &flags))
604 return 0;
605 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
606 unlock_task_sighand(task, &flags);
607
608 /*
609 * print the file header
610 */
611 seq_puts(m, "Limit "
612 "Soft Limit "
613 "Hard Limit "
614 "Units \n");
615
616 for (i = 0; i < RLIM_NLIMITS; i++) {
617 if (rlim[i].rlim_cur == RLIM_INFINITY)
618 seq_printf(m, "%-25s %-20s ",
619 lnames[i].name, "unlimited");
620 else
621 seq_printf(m, "%-25s %-20lu ",
622 lnames[i].name, rlim[i].rlim_cur);
623
624 if (rlim[i].rlim_max == RLIM_INFINITY)
625 seq_printf(m, "%-20s ", "unlimited");
626 else
627 seq_printf(m, "%-20lu ", rlim[i].rlim_max);
628
629 if (lnames[i].unit)
630 seq_printf(m, "%-10s\n", lnames[i].unit);
631 else
632 seq_putc(m, '\n');
633 }
634
635 return 0;
636}
637
638#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
639static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
640 struct pid *pid, struct task_struct *task)
641{
642 struct syscall_info info;
643 u64 *args = &info.data.args[0];
644 int res;
645
646 res = lock_trace(task);
647 if (res)
648 return res;
649
650 if (task_current_syscall(task, &info))
651 seq_puts(m, "running\n");
652 else if (info.data.nr < 0)
653 seq_printf(m, "%d 0x%llx 0x%llx\n",
654 info.data.nr, info.sp, info.data.instruction_pointer);
655 else
656 seq_printf(m,
657 "%d 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx\n",
658 info.data.nr,
659 args[0], args[1], args[2], args[3], args[4], args[5],
660 info.sp, info.data.instruction_pointer);
661 unlock_trace(task);
662
663 return 0;
664}
665#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
666
667/************************************************************************/
668/* Here the fs part begins */
669/************************************************************************/
670
671/* permission checks */
672static int proc_fd_access_allowed(struct inode *inode)
673{
674 struct task_struct *task;
675 int allowed = 0;
676 /* Allow access to a task's file descriptors if it is us or we
677 * may use ptrace attach to the process and find out that
678 * information.
679 */
680 task = get_proc_task(inode);
681 if (task) {
682 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
683 put_task_struct(task);
684 }
685 return allowed;
686}
687
688int proc_setattr(struct dentry *dentry, struct iattr *attr)
689{
690 int error;
691 struct inode *inode = d_inode(dentry);
692
693 if (attr->ia_valid & ATTR_MODE)
694 return -EPERM;
695
696 error = setattr_prepare(dentry, attr);
697 if (error)
698 return error;
699
700 setattr_copy(inode, attr);
701 mark_inode_dirty(inode);
702 return 0;
703}
704
705/*
706 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
707 * or euid/egid (for hide_pid_min=2)?
708 */
709static bool has_pid_permissions(struct proc_fs_info *fs_info,
710 struct task_struct *task,
711 enum proc_hidepid hide_pid_min)
712{
713 /*
714 * If 'hidpid' mount option is set force a ptrace check,
715 * we indicate that we are using a filesystem syscall
716 * by passing PTRACE_MODE_READ_FSCREDS
717 */
718 if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE)
719 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
720
721 if (fs_info->hide_pid < hide_pid_min)
722 return true;
723 if (in_group_p(fs_info->pid_gid))
724 return true;
725 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
726}
727
728
729static int proc_pid_permission(struct inode *inode, int mask)
730{
731 struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
732 struct task_struct *task;
733 bool has_perms;
734
735 task = get_proc_task(inode);
736 if (!task)
737 return -ESRCH;
738 has_perms = has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS);
739 put_task_struct(task);
740
741 if (!has_perms) {
742 if (fs_info->hide_pid == HIDEPID_INVISIBLE) {
743 /*
744 * Let's make getdents(), stat(), and open()
745 * consistent with each other. If a process
746 * may not stat() a file, it shouldn't be seen
747 * in procfs at all.
748 */
749 return -ENOENT;
750 }
751
752 return -EPERM;
753 }
754 return generic_permission(inode, mask);
755}
756
757
758
759static const struct inode_operations proc_def_inode_operations = {
760 .setattr = proc_setattr,
761};
762
763static int proc_single_show(struct seq_file *m, void *v)
764{
765 struct inode *inode = m->private;
766 struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
767 struct pid *pid = proc_pid(inode);
768 struct task_struct *task;
769 int ret;
770
771 task = get_pid_task(pid, PIDTYPE_PID);
772 if (!task)
773 return -ESRCH;
774
775 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
776
777 put_task_struct(task);
778 return ret;
779}
780
781static int proc_single_open(struct inode *inode, struct file *filp)
782{
783 return single_open(filp, proc_single_show, inode);
784}
785
786static const struct file_operations proc_single_file_operations = {
787 .open = proc_single_open,
788 .read = seq_read,
789 .llseek = seq_lseek,
790 .release = single_release,
791};
792
793
794struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
795{
796 struct task_struct *task = get_proc_task(inode);
797 struct mm_struct *mm = ERR_PTR(-ESRCH);
798
799 if (task) {
800 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
801 put_task_struct(task);
802
803 if (!IS_ERR_OR_NULL(mm)) {
804 /* ensure this mm_struct can't be freed */
805 mmgrab(mm);
806 /* but do not pin its memory */
807 mmput(mm);
808 }
809 }
810
811 return mm;
812}
813
814static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
815{
816 struct mm_struct *mm = proc_mem_open(inode, mode);
817
818 if (IS_ERR(mm))
819 return PTR_ERR(mm);
820
821 file->private_data = mm;
822 return 0;
823}
824
825static int mem_open(struct inode *inode, struct file *file)
826{
827 int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
828
829 /* OK to pass negative loff_t, we can catch out-of-range */
830 file->f_mode |= FMODE_UNSIGNED_OFFSET;
831
832 return ret;
833}
834
835static ssize_t mem_rw(struct file *file, char __user *buf,
836 size_t count, loff_t *ppos, int write)
837{
838 struct mm_struct *mm = file->private_data;
839 unsigned long addr = *ppos;
840 ssize_t copied;
841 char *page;
842 unsigned int flags;
843
844 if (!mm)
845 return 0;
846
847 page = (char *)__get_free_page(GFP_KERNEL);
848 if (!page)
849 return -ENOMEM;
850
851 copied = 0;
852 if (!mmget_not_zero(mm))
853 goto free;
854
855 flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
856
857 while (count > 0) {
858 int this_len = min_t(int, count, PAGE_SIZE);
859
860 if (write && copy_from_user(page, buf, this_len)) {
861 copied = -EFAULT;
862 break;
863 }
864
865 this_len = access_remote_vm(mm, addr, page, this_len, flags);
866 if (!this_len) {
867 if (!copied)
868 copied = -EIO;
869 break;
870 }
871
872 if (!write && copy_to_user(buf, page, this_len)) {
873 copied = -EFAULT;
874 break;
875 }
876
877 buf += this_len;
878 addr += this_len;
879 copied += this_len;
880 count -= this_len;
881 }
882 *ppos = addr;
883
884 mmput(mm);
885free:
886 free_page((unsigned long) page);
887 return copied;
888}
889
890static ssize_t mem_read(struct file *file, char __user *buf,
891 size_t count, loff_t *ppos)
892{
893 return mem_rw(file, buf, count, ppos, 0);
894}
895
896static ssize_t mem_write(struct file *file, const char __user *buf,
897 size_t count, loff_t *ppos)
898{
899 return mem_rw(file, (char __user*)buf, count, ppos, 1);
900}
901
902loff_t mem_lseek(struct file *file, loff_t offset, int orig)
903{
904 switch (orig) {
905 case 0:
906 file->f_pos = offset;
907 break;
908 case 1:
909 file->f_pos += offset;
910 break;
911 default:
912 return -EINVAL;
913 }
914 force_successful_syscall_return();
915 return file->f_pos;
916}
917
918static int mem_release(struct inode *inode, struct file *file)
919{
920 struct mm_struct *mm = file->private_data;
921 if (mm)
922 mmdrop(mm);
923 return 0;
924}
925
926static const struct file_operations proc_mem_operations = {
927 .llseek = mem_lseek,
928 .read = mem_read,
929 .write = mem_write,
930 .open = mem_open,
931 .release = mem_release,
932};
933
934static int environ_open(struct inode *inode, struct file *file)
935{
936 return __mem_open(inode, file, PTRACE_MODE_READ);
937}
938
939static ssize_t environ_read(struct file *file, char __user *buf,
940 size_t count, loff_t *ppos)
941{
942 char *page;
943 unsigned long src = *ppos;
944 int ret = 0;
945 struct mm_struct *mm = file->private_data;
946 unsigned long env_start, env_end;
947
948 /* Ensure the process spawned far enough to have an environment. */
949 if (!mm || !mm->env_end)
950 return 0;
951
952 page = (char *)__get_free_page(GFP_KERNEL);
953 if (!page)
954 return -ENOMEM;
955
956 ret = 0;
957 if (!mmget_not_zero(mm))
958 goto free;
959
960 spin_lock(&mm->arg_lock);
961 env_start = mm->env_start;
962 env_end = mm->env_end;
963 spin_unlock(&mm->arg_lock);
964
965 while (count > 0) {
966 size_t this_len, max_len;
967 int retval;
968
969 if (src >= (env_end - env_start))
970 break;
971
972 this_len = env_end - (env_start + src);
973
974 max_len = min_t(size_t, PAGE_SIZE, count);
975 this_len = min(max_len, this_len);
976
977 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
978
979 if (retval <= 0) {
980 ret = retval;
981 break;
982 }
983
984 if (copy_to_user(buf, page, retval)) {
985 ret = -EFAULT;
986 break;
987 }
988
989 ret += retval;
990 src += retval;
991 buf += retval;
992 count -= retval;
993 }
994 *ppos = src;
995 mmput(mm);
996
997free:
998 free_page((unsigned long) page);
999 return ret;
1000}
1001
1002static const struct file_operations proc_environ_operations = {
1003 .open = environ_open,
1004 .read = environ_read,
1005 .llseek = generic_file_llseek,
1006 .release = mem_release,
1007};
1008
1009static int auxv_open(struct inode *inode, struct file *file)
1010{
1011 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1012}
1013
1014static ssize_t auxv_read(struct file *file, char __user *buf,
1015 size_t count, loff_t *ppos)
1016{
1017 struct mm_struct *mm = file->private_data;
1018 unsigned int nwords = 0;
1019
1020 if (!mm)
1021 return 0;
1022 do {
1023 nwords += 2;
1024 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1025 return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1026 nwords * sizeof(mm->saved_auxv[0]));
1027}
1028
1029static const struct file_operations proc_auxv_operations = {
1030 .open = auxv_open,
1031 .read = auxv_read,
1032 .llseek = generic_file_llseek,
1033 .release = mem_release,
1034};
1035
1036static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1037 loff_t *ppos)
1038{
1039 struct task_struct *task = get_proc_task(file_inode(file));
1040 char buffer[PROC_NUMBUF];
1041 int oom_adj = OOM_ADJUST_MIN;
1042 size_t len;
1043
1044 if (!task)
1045 return -ESRCH;
1046 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1047 oom_adj = OOM_ADJUST_MAX;
1048 else
1049 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1050 OOM_SCORE_ADJ_MAX;
1051 put_task_struct(task);
1052 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1053 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1054}
1055
1056static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1057{
1058 static DEFINE_MUTEX(oom_adj_mutex);
1059 struct mm_struct *mm = NULL;
1060 struct task_struct *task;
1061 int err = 0;
1062
1063 task = get_proc_task(file_inode(file));
1064 if (!task)
1065 return -ESRCH;
1066
1067 mutex_lock(&oom_adj_mutex);
1068 if (legacy) {
1069 if (oom_adj < task->signal->oom_score_adj &&
1070 !capable(CAP_SYS_RESOURCE)) {
1071 err = -EACCES;
1072 goto err_unlock;
1073 }
1074 /*
1075 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1076 * /proc/pid/oom_score_adj instead.
1077 */
1078 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1079 current->comm, task_pid_nr(current), task_pid_nr(task),
1080 task_pid_nr(task));
1081 } else {
1082 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1083 !capable(CAP_SYS_RESOURCE)) {
1084 err = -EACCES;
1085 goto err_unlock;
1086 }
1087 }
1088
1089 /*
1090 * Make sure we will check other processes sharing the mm if this is
1091 * not vfrok which wants its own oom_score_adj.
1092 * pin the mm so it doesn't go away and get reused after task_unlock
1093 */
1094 if (!task->vfork_done) {
1095 struct task_struct *p = find_lock_task_mm(task);
1096
1097 if (p) {
1098 if (atomic_read(&p->mm->mm_users) > 1) {
1099 mm = p->mm;
1100 mmgrab(mm);
1101 }
1102 task_unlock(p);
1103 }
1104 }
1105
1106 task->signal->oom_score_adj = oom_adj;
1107 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1108 task->signal->oom_score_adj_min = (short)oom_adj;
1109 trace_oom_score_adj_update(task);
1110
1111 if (mm) {
1112 struct task_struct *p;
1113
1114 rcu_read_lock();
1115 for_each_process(p) {
1116 if (same_thread_group(task, p))
1117 continue;
1118
1119 /* do not touch kernel threads or the global init */
1120 if (p->flags & PF_KTHREAD || is_global_init(p))
1121 continue;
1122
1123 task_lock(p);
1124 if (!p->vfork_done && process_shares_mm(p, mm)) {
1125 p->signal->oom_score_adj = oom_adj;
1126 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1127 p->signal->oom_score_adj_min = (short)oom_adj;
1128 }
1129 task_unlock(p);
1130 }
1131 rcu_read_unlock();
1132 mmdrop(mm);
1133 }
1134err_unlock:
1135 mutex_unlock(&oom_adj_mutex);
1136 put_task_struct(task);
1137 return err;
1138}
1139
1140/*
1141 * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1142 * kernels. The effective policy is defined by oom_score_adj, which has a
1143 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1144 * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1145 * Processes that become oom disabled via oom_adj will still be oom disabled
1146 * with this implementation.
1147 *
1148 * oom_adj cannot be removed since existing userspace binaries use it.
1149 */
1150static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1151 size_t count, loff_t *ppos)
1152{
1153 char buffer[PROC_NUMBUF];
1154 int oom_adj;
1155 int err;
1156
1157 memset(buffer, 0, sizeof(buffer));
1158 if (count > sizeof(buffer) - 1)
1159 count = sizeof(buffer) - 1;
1160 if (copy_from_user(buffer, buf, count)) {
1161 err = -EFAULT;
1162 goto out;
1163 }
1164
1165 err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1166 if (err)
1167 goto out;
1168 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1169 oom_adj != OOM_DISABLE) {
1170 err = -EINVAL;
1171 goto out;
1172 }
1173
1174 /*
1175 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1176 * value is always attainable.
1177 */
1178 if (oom_adj == OOM_ADJUST_MAX)
1179 oom_adj = OOM_SCORE_ADJ_MAX;
1180 else
1181 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1182
1183 err = __set_oom_adj(file, oom_adj, true);
1184out:
1185 return err < 0 ? err : count;
1186}
1187
1188static const struct file_operations proc_oom_adj_operations = {
1189 .read = oom_adj_read,
1190 .write = oom_adj_write,
1191 .llseek = generic_file_llseek,
1192};
1193
1194static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1195 size_t count, loff_t *ppos)
1196{
1197 struct task_struct *task = get_proc_task(file_inode(file));
1198 char buffer[PROC_NUMBUF];
1199 short oom_score_adj = OOM_SCORE_ADJ_MIN;
1200 size_t len;
1201
1202 if (!task)
1203 return -ESRCH;
1204 oom_score_adj = task->signal->oom_score_adj;
1205 put_task_struct(task);
1206 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1207 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1208}
1209
1210static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1211 size_t count, loff_t *ppos)
1212{
1213 char buffer[PROC_NUMBUF];
1214 int oom_score_adj;
1215 int err;
1216
1217 memset(buffer, 0, sizeof(buffer));
1218 if (count > sizeof(buffer) - 1)
1219 count = sizeof(buffer) - 1;
1220 if (copy_from_user(buffer, buf, count)) {
1221 err = -EFAULT;
1222 goto out;
1223 }
1224
1225 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1226 if (err)
1227 goto out;
1228 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1229 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1230 err = -EINVAL;
1231 goto out;
1232 }
1233
1234 err = __set_oom_adj(file, oom_score_adj, false);
1235out:
1236 return err < 0 ? err : count;
1237}
1238
1239static const struct file_operations proc_oom_score_adj_operations = {
1240 .read = oom_score_adj_read,
1241 .write = oom_score_adj_write,
1242 .llseek = default_llseek,
1243};
1244
1245#ifdef CONFIG_AUDIT
1246#define TMPBUFLEN 11
1247static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1248 size_t count, loff_t *ppos)
1249{
1250 struct inode * inode = file_inode(file);
1251 struct task_struct *task = get_proc_task(inode);
1252 ssize_t length;
1253 char tmpbuf[TMPBUFLEN];
1254
1255 if (!task)
1256 return -ESRCH;
1257 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1258 from_kuid(file->f_cred->user_ns,
1259 audit_get_loginuid(task)));
1260 put_task_struct(task);
1261 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1262}
1263
1264static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1265 size_t count, loff_t *ppos)
1266{
1267 struct inode * inode = file_inode(file);
1268 uid_t loginuid;
1269 kuid_t kloginuid;
1270 int rv;
1271
1272 rcu_read_lock();
1273 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1274 rcu_read_unlock();
1275 return -EPERM;
1276 }
1277 rcu_read_unlock();
1278
1279 if (*ppos != 0) {
1280 /* No partial writes. */
1281 return -EINVAL;
1282 }
1283
1284 rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1285 if (rv < 0)
1286 return rv;
1287
1288 /* is userspace tring to explicitly UNSET the loginuid? */
1289 if (loginuid == AUDIT_UID_UNSET) {
1290 kloginuid = INVALID_UID;
1291 } else {
1292 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1293 if (!uid_valid(kloginuid))
1294 return -EINVAL;
1295 }
1296
1297 rv = audit_set_loginuid(kloginuid);
1298 if (rv < 0)
1299 return rv;
1300 return count;
1301}
1302
1303static const struct file_operations proc_loginuid_operations = {
1304 .read = proc_loginuid_read,
1305 .write = proc_loginuid_write,
1306 .llseek = generic_file_llseek,
1307};
1308
1309static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1310 size_t count, loff_t *ppos)
1311{
1312 struct inode * inode = file_inode(file);
1313 struct task_struct *task = get_proc_task(inode);
1314 ssize_t length;
1315 char tmpbuf[TMPBUFLEN];
1316
1317 if (!task)
1318 return -ESRCH;
1319 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1320 audit_get_sessionid(task));
1321 put_task_struct(task);
1322 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1323}
1324
1325static const struct file_operations proc_sessionid_operations = {
1326 .read = proc_sessionid_read,
1327 .llseek = generic_file_llseek,
1328};
1329#endif
1330
1331#ifdef CONFIG_FAULT_INJECTION
1332static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1333 size_t count, loff_t *ppos)
1334{
1335 struct task_struct *task = get_proc_task(file_inode(file));
1336 char buffer[PROC_NUMBUF];
1337 size_t len;
1338 int make_it_fail;
1339
1340 if (!task)
1341 return -ESRCH;
1342 make_it_fail = task->make_it_fail;
1343 put_task_struct(task);
1344
1345 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1346
1347 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1348}
1349
1350static ssize_t proc_fault_inject_write(struct file * file,
1351 const char __user * buf, size_t count, loff_t *ppos)
1352{
1353 struct task_struct *task;
1354 char buffer[PROC_NUMBUF];
1355 int make_it_fail;
1356 int rv;
1357
1358 if (!capable(CAP_SYS_RESOURCE))
1359 return -EPERM;
1360 memset(buffer, 0, sizeof(buffer));
1361 if (count > sizeof(buffer) - 1)
1362 count = sizeof(buffer) - 1;
1363 if (copy_from_user(buffer, buf, count))
1364 return -EFAULT;
1365 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1366 if (rv < 0)
1367 return rv;
1368 if (make_it_fail < 0 || make_it_fail > 1)
1369 return -EINVAL;
1370
1371 task = get_proc_task(file_inode(file));
1372 if (!task)
1373 return -ESRCH;
1374 task->make_it_fail = make_it_fail;
1375 put_task_struct(task);
1376
1377 return count;
1378}
1379
1380static const struct file_operations proc_fault_inject_operations = {
1381 .read = proc_fault_inject_read,
1382 .write = proc_fault_inject_write,
1383 .llseek = generic_file_llseek,
1384};
1385
1386static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1387 size_t count, loff_t *ppos)
1388{
1389 struct task_struct *task;
1390 int err;
1391 unsigned int n;
1392
1393 err = kstrtouint_from_user(buf, count, 0, &n);
1394 if (err)
1395 return err;
1396
1397 task = get_proc_task(file_inode(file));
1398 if (!task)
1399 return -ESRCH;
1400 task->fail_nth = n;
1401 put_task_struct(task);
1402
1403 return count;
1404}
1405
1406static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1407 size_t count, loff_t *ppos)
1408{
1409 struct task_struct *task;
1410 char numbuf[PROC_NUMBUF];
1411 ssize_t len;
1412
1413 task = get_proc_task(file_inode(file));
1414 if (!task)
1415 return -ESRCH;
1416 len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1417 put_task_struct(task);
1418 return simple_read_from_buffer(buf, count, ppos, numbuf, len);
1419}
1420
1421static const struct file_operations proc_fail_nth_operations = {
1422 .read = proc_fail_nth_read,
1423 .write = proc_fail_nth_write,
1424};
1425#endif
1426
1427
1428#ifdef CONFIG_SCHED_DEBUG
1429/*
1430 * Print out various scheduling related per-task fields:
1431 */
1432static int sched_show(struct seq_file *m, void *v)
1433{
1434 struct inode *inode = m->private;
1435 struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
1436 struct task_struct *p;
1437
1438 p = get_proc_task(inode);
1439 if (!p)
1440 return -ESRCH;
1441 proc_sched_show_task(p, ns, m);
1442
1443 put_task_struct(p);
1444
1445 return 0;
1446}
1447
1448static ssize_t
1449sched_write(struct file *file, const char __user *buf,
1450 size_t count, loff_t *offset)
1451{
1452 struct inode *inode = file_inode(file);
1453 struct task_struct *p;
1454
1455 p = get_proc_task(inode);
1456 if (!p)
1457 return -ESRCH;
1458 proc_sched_set_task(p);
1459
1460 put_task_struct(p);
1461
1462 return count;
1463}
1464
1465static int sched_open(struct inode *inode, struct file *filp)
1466{
1467 return single_open(filp, sched_show, inode);
1468}
1469
1470static const struct file_operations proc_pid_sched_operations = {
1471 .open = sched_open,
1472 .read = seq_read,
1473 .write = sched_write,
1474 .llseek = seq_lseek,
1475 .release = single_release,
1476};
1477
1478#endif
1479
1480#ifdef CONFIG_SCHED_AUTOGROUP
1481/*
1482 * Print out autogroup related information:
1483 */
1484static int sched_autogroup_show(struct seq_file *m, void *v)
1485{
1486 struct inode *inode = m->private;
1487 struct task_struct *p;
1488
1489 p = get_proc_task(inode);
1490 if (!p)
1491 return -ESRCH;
1492 proc_sched_autogroup_show_task(p, m);
1493
1494 put_task_struct(p);
1495
1496 return 0;
1497}
1498
1499static ssize_t
1500sched_autogroup_write(struct file *file, const char __user *buf,
1501 size_t count, loff_t *offset)
1502{
1503 struct inode *inode = file_inode(file);
1504 struct task_struct *p;
1505 char buffer[PROC_NUMBUF];
1506 int nice;
1507 int err;
1508
1509 memset(buffer, 0, sizeof(buffer));
1510 if (count > sizeof(buffer) - 1)
1511 count = sizeof(buffer) - 1;
1512 if (copy_from_user(buffer, buf, count))
1513 return -EFAULT;
1514
1515 err = kstrtoint(strstrip(buffer), 0, &nice);
1516 if (err < 0)
1517 return err;
1518
1519 p = get_proc_task(inode);
1520 if (!p)
1521 return -ESRCH;
1522
1523 err = proc_sched_autogroup_set_nice(p, nice);
1524 if (err)
1525 count = err;
1526
1527 put_task_struct(p);
1528
1529 return count;
1530}
1531
1532static int sched_autogroup_open(struct inode *inode, struct file *filp)
1533{
1534 int ret;
1535
1536 ret = single_open(filp, sched_autogroup_show, NULL);
1537 if (!ret) {
1538 struct seq_file *m = filp->private_data;
1539
1540 m->private = inode;
1541 }
1542 return ret;
1543}
1544
1545static const struct file_operations proc_pid_sched_autogroup_operations = {
1546 .open = sched_autogroup_open,
1547 .read = seq_read,
1548 .write = sched_autogroup_write,
1549 .llseek = seq_lseek,
1550 .release = single_release,
1551};
1552
1553#endif /* CONFIG_SCHED_AUTOGROUP */
1554
1555#ifdef CONFIG_TIME_NS
1556static int timens_offsets_show(struct seq_file *m, void *v)
1557{
1558 struct task_struct *p;
1559
1560 p = get_proc_task(file_inode(m->file));
1561 if (!p)
1562 return -ESRCH;
1563 proc_timens_show_offsets(p, m);
1564
1565 put_task_struct(p);
1566
1567 return 0;
1568}
1569
1570static ssize_t timens_offsets_write(struct file *file, const char __user *buf,
1571 size_t count, loff_t *ppos)
1572{
1573 struct inode *inode = file_inode(file);
1574 struct proc_timens_offset offsets[2];
1575 char *kbuf = NULL, *pos, *next_line;
1576 struct task_struct *p;
1577 int ret, noffsets;
1578
1579 /* Only allow < page size writes at the beginning of the file */
1580 if ((*ppos != 0) || (count >= PAGE_SIZE))
1581 return -EINVAL;
1582
1583 /* Slurp in the user data */
1584 kbuf = memdup_user_nul(buf, count);
1585 if (IS_ERR(kbuf))
1586 return PTR_ERR(kbuf);
1587
1588 /* Parse the user data */
1589 ret = -EINVAL;
1590 noffsets = 0;
1591 for (pos = kbuf; pos; pos = next_line) {
1592 struct proc_timens_offset *off = &offsets[noffsets];
1593 char clock[10];
1594 int err;
1595
1596 /* Find the end of line and ensure we don't look past it */
1597 next_line = strchr(pos, '\n');
1598 if (next_line) {
1599 *next_line = '\0';
1600 next_line++;
1601 if (*next_line == '\0')
1602 next_line = NULL;
1603 }
1604
1605 err = sscanf(pos, "%9s %lld %lu", clock,
1606 &off->val.tv_sec, &off->val.tv_nsec);
1607 if (err != 3 || off->val.tv_nsec >= NSEC_PER_SEC)
1608 goto out;
1609
1610 clock[sizeof(clock) - 1] = 0;
1611 if (strcmp(clock, "monotonic") == 0 ||
1612 strcmp(clock, __stringify(CLOCK_MONOTONIC)) == 0)
1613 off->clockid = CLOCK_MONOTONIC;
1614 else if (strcmp(clock, "boottime") == 0 ||
1615 strcmp(clock, __stringify(CLOCK_BOOTTIME)) == 0)
1616 off->clockid = CLOCK_BOOTTIME;
1617 else
1618 goto out;
1619
1620 noffsets++;
1621 if (noffsets == ARRAY_SIZE(offsets)) {
1622 if (next_line)
1623 count = next_line - kbuf;
1624 break;
1625 }
1626 }
1627
1628 ret = -ESRCH;
1629 p = get_proc_task(inode);
1630 if (!p)
1631 goto out;
1632 ret = proc_timens_set_offset(file, p, offsets, noffsets);
1633 put_task_struct(p);
1634 if (ret)
1635 goto out;
1636
1637 ret = count;
1638out:
1639 kfree(kbuf);
1640 return ret;
1641}
1642
1643static int timens_offsets_open(struct inode *inode, struct file *filp)
1644{
1645 return single_open(filp, timens_offsets_show, inode);
1646}
1647
1648static const struct file_operations proc_timens_offsets_operations = {
1649 .open = timens_offsets_open,
1650 .read = seq_read,
1651 .write = timens_offsets_write,
1652 .llseek = seq_lseek,
1653 .release = single_release,
1654};
1655#endif /* CONFIG_TIME_NS */
1656
1657static ssize_t comm_write(struct file *file, const char __user *buf,
1658 size_t count, loff_t *offset)
1659{
1660 struct inode *inode = file_inode(file);
1661 struct task_struct *p;
1662 char buffer[TASK_COMM_LEN];
1663 const size_t maxlen = sizeof(buffer) - 1;
1664
1665 memset(buffer, 0, sizeof(buffer));
1666 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1667 return -EFAULT;
1668
1669 p = get_proc_task(inode);
1670 if (!p)
1671 return -ESRCH;
1672
1673 if (same_thread_group(current, p))
1674 set_task_comm(p, buffer);
1675 else
1676 count = -EINVAL;
1677
1678 put_task_struct(p);
1679
1680 return count;
1681}
1682
1683static int comm_show(struct seq_file *m, void *v)
1684{
1685 struct inode *inode = m->private;
1686 struct task_struct *p;
1687
1688 p = get_proc_task(inode);
1689 if (!p)
1690 return -ESRCH;
1691
1692 proc_task_name(m, p, false);
1693 seq_putc(m, '\n');
1694
1695 put_task_struct(p);
1696
1697 return 0;
1698}
1699
1700static int comm_open(struct inode *inode, struct file *filp)
1701{
1702 return single_open(filp, comm_show, inode);
1703}
1704
1705static const struct file_operations proc_pid_set_comm_operations = {
1706 .open = comm_open,
1707 .read = seq_read,
1708 .write = comm_write,
1709 .llseek = seq_lseek,
1710 .release = single_release,
1711};
1712
1713static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1714{
1715 struct task_struct *task;
1716 struct file *exe_file;
1717
1718 task = get_proc_task(d_inode(dentry));
1719 if (!task)
1720 return -ENOENT;
1721 exe_file = get_task_exe_file(task);
1722 put_task_struct(task);
1723 if (exe_file) {
1724 *exe_path = exe_file->f_path;
1725 path_get(&exe_file->f_path);
1726 fput(exe_file);
1727 return 0;
1728 } else
1729 return -ENOENT;
1730}
1731
1732static const char *proc_pid_get_link(struct dentry *dentry,
1733 struct inode *inode,
1734 struct delayed_call *done)
1735{
1736 struct path path;
1737 int error = -EACCES;
1738
1739 if (!dentry)
1740 return ERR_PTR(-ECHILD);
1741
1742 /* Are we allowed to snoop on the tasks file descriptors? */
1743 if (!proc_fd_access_allowed(inode))
1744 goto out;
1745
1746 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1747 if (error)
1748 goto out;
1749
1750 error = nd_jump_link(&path);
1751out:
1752 return ERR_PTR(error);
1753}
1754
1755static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1756{
1757 char *tmp = (char *)__get_free_page(GFP_KERNEL);
1758 char *pathname;
1759 int len;
1760
1761 if (!tmp)
1762 return -ENOMEM;
1763
1764 pathname = d_path(path, tmp, PAGE_SIZE);
1765 len = PTR_ERR(pathname);
1766 if (IS_ERR(pathname))
1767 goto out;
1768 len = tmp + PAGE_SIZE - 1 - pathname;
1769
1770 if (len > buflen)
1771 len = buflen;
1772 if (copy_to_user(buffer, pathname, len))
1773 len = -EFAULT;
1774 out:
1775 free_page((unsigned long)tmp);
1776 return len;
1777}
1778
1779static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1780{
1781 int error = -EACCES;
1782 struct inode *inode = d_inode(dentry);
1783 struct path path;
1784
1785 /* Are we allowed to snoop on the tasks file descriptors? */
1786 if (!proc_fd_access_allowed(inode))
1787 goto out;
1788
1789 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1790 if (error)
1791 goto out;
1792
1793 error = do_proc_readlink(&path, buffer, buflen);
1794 path_put(&path);
1795out:
1796 return error;
1797}
1798
1799const struct inode_operations proc_pid_link_inode_operations = {
1800 .readlink = proc_pid_readlink,
1801 .get_link = proc_pid_get_link,
1802 .setattr = proc_setattr,
1803};
1804
1805
1806/* building an inode */
1807
1808void task_dump_owner(struct task_struct *task, umode_t mode,
1809 kuid_t *ruid, kgid_t *rgid)
1810{
1811 /* Depending on the state of dumpable compute who should own a
1812 * proc file for a task.
1813 */
1814 const struct cred *cred;
1815 kuid_t uid;
1816 kgid_t gid;
1817
1818 if (unlikely(task->flags & PF_KTHREAD)) {
1819 *ruid = GLOBAL_ROOT_UID;
1820 *rgid = GLOBAL_ROOT_GID;
1821 return;
1822 }
1823
1824 /* Default to the tasks effective ownership */
1825 rcu_read_lock();
1826 cred = __task_cred(task);
1827 uid = cred->euid;
1828 gid = cred->egid;
1829 rcu_read_unlock();
1830
1831 /*
1832 * Before the /proc/pid/status file was created the only way to read
1833 * the effective uid of a /process was to stat /proc/pid. Reading
1834 * /proc/pid/status is slow enough that procps and other packages
1835 * kept stating /proc/pid. To keep the rules in /proc simple I have
1836 * made this apply to all per process world readable and executable
1837 * directories.
1838 */
1839 if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1840 struct mm_struct *mm;
1841 task_lock(task);
1842 mm = task->mm;
1843 /* Make non-dumpable tasks owned by some root */
1844 if (mm) {
1845 if (get_dumpable(mm) != SUID_DUMP_USER) {
1846 struct user_namespace *user_ns = mm->user_ns;
1847
1848 uid = make_kuid(user_ns, 0);
1849 if (!uid_valid(uid))
1850 uid = GLOBAL_ROOT_UID;
1851
1852 gid = make_kgid(user_ns, 0);
1853 if (!gid_valid(gid))
1854 gid = GLOBAL_ROOT_GID;
1855 }
1856 } else {
1857 uid = GLOBAL_ROOT_UID;
1858 gid = GLOBAL_ROOT_GID;
1859 }
1860 task_unlock(task);
1861 }
1862 *ruid = uid;
1863 *rgid = gid;
1864}
1865
1866void proc_pid_evict_inode(struct proc_inode *ei)
1867{
1868 struct pid *pid = ei->pid;
1869
1870 if (S_ISDIR(ei->vfs_inode.i_mode)) {
1871 spin_lock(&pid->lock);
1872 hlist_del_init_rcu(&ei->sibling_inodes);
1873 spin_unlock(&pid->lock);
1874 }
1875
1876 put_pid(pid);
1877}
1878
1879struct inode *proc_pid_make_inode(struct super_block * sb,
1880 struct task_struct *task, umode_t mode)
1881{
1882 struct inode * inode;
1883 struct proc_inode *ei;
1884 struct pid *pid;
1885
1886 /* We need a new inode */
1887
1888 inode = new_inode(sb);
1889 if (!inode)
1890 goto out;
1891
1892 /* Common stuff */
1893 ei = PROC_I(inode);
1894 inode->i_mode = mode;
1895 inode->i_ino = get_next_ino();
1896 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1897 inode->i_op = &proc_def_inode_operations;
1898
1899 /*
1900 * grab the reference to task.
1901 */
1902 pid = get_task_pid(task, PIDTYPE_PID);
1903 if (!pid)
1904 goto out_unlock;
1905
1906 /* Let the pid remember us for quick removal */
1907 ei->pid = pid;
1908 if (S_ISDIR(mode)) {
1909 spin_lock(&pid->lock);
1910 hlist_add_head_rcu(&ei->sibling_inodes, &pid->inodes);
1911 spin_unlock(&pid->lock);
1912 }
1913
1914 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1915 security_task_to_inode(task, inode);
1916
1917out:
1918 return inode;
1919
1920out_unlock:
1921 iput(inode);
1922 return NULL;
1923}
1924
1925int pid_getattr(const struct path *path, struct kstat *stat,
1926 u32 request_mask, unsigned int query_flags)
1927{
1928 struct inode *inode = d_inode(path->dentry);
1929 struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
1930 struct task_struct *task;
1931
1932 generic_fillattr(inode, stat);
1933
1934 stat->uid = GLOBAL_ROOT_UID;
1935 stat->gid = GLOBAL_ROOT_GID;
1936 rcu_read_lock();
1937 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1938 if (task) {
1939 if (!has_pid_permissions(fs_info, task, HIDEPID_INVISIBLE)) {
1940 rcu_read_unlock();
1941 /*
1942 * This doesn't prevent learning whether PID exists,
1943 * it only makes getattr() consistent with readdir().
1944 */
1945 return -ENOENT;
1946 }
1947 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1948 }
1949 rcu_read_unlock();
1950 return 0;
1951}
1952
1953/* dentry stuff */
1954
1955/*
1956 * Set <pid>/... inode ownership (can change due to setuid(), etc.)
1957 */
1958void pid_update_inode(struct task_struct *task, struct inode *inode)
1959{
1960 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1961
1962 inode->i_mode &= ~(S_ISUID | S_ISGID);
1963 security_task_to_inode(task, inode);
1964}
1965
1966/*
1967 * Rewrite the inode's ownerships here because the owning task may have
1968 * performed a setuid(), etc.
1969 *
1970 */
1971static int pid_revalidate(struct dentry *dentry, unsigned int flags)
1972{
1973 struct inode *inode;
1974 struct task_struct *task;
1975
1976 if (flags & LOOKUP_RCU)
1977 return -ECHILD;
1978
1979 inode = d_inode(dentry);
1980 task = get_proc_task(inode);
1981
1982 if (task) {
1983 pid_update_inode(task, inode);
1984 put_task_struct(task);
1985 return 1;
1986 }
1987 return 0;
1988}
1989
1990static inline bool proc_inode_is_dead(struct inode *inode)
1991{
1992 return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1993}
1994
1995int pid_delete_dentry(const struct dentry *dentry)
1996{
1997 /* Is the task we represent dead?
1998 * If so, then don't put the dentry on the lru list,
1999 * kill it immediately.
2000 */
2001 return proc_inode_is_dead(d_inode(dentry));
2002}
2003
2004const struct dentry_operations pid_dentry_operations =
2005{
2006 .d_revalidate = pid_revalidate,
2007 .d_delete = pid_delete_dentry,
2008};
2009
2010/* Lookups */
2011
2012/*
2013 * Fill a directory entry.
2014 *
2015 * If possible create the dcache entry and derive our inode number and
2016 * file type from dcache entry.
2017 *
2018 * Since all of the proc inode numbers are dynamically generated, the inode
2019 * numbers do not exist until the inode is cache. This means creating the
2020 * the dcache entry in readdir is necessary to keep the inode numbers
2021 * reported by readdir in sync with the inode numbers reported
2022 * by stat.
2023 */
2024bool proc_fill_cache(struct file *file, struct dir_context *ctx,
2025 const char *name, unsigned int len,
2026 instantiate_t instantiate, struct task_struct *task, const void *ptr)
2027{
2028 struct dentry *child, *dir = file->f_path.dentry;
2029 struct qstr qname = QSTR_INIT(name, len);
2030 struct inode *inode;
2031 unsigned type = DT_UNKNOWN;
2032 ino_t ino = 1;
2033
2034 child = d_hash_and_lookup(dir, &qname);
2035 if (!child) {
2036 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2037 child = d_alloc_parallel(dir, &qname, &wq);
2038 if (IS_ERR(child))
2039 goto end_instantiate;
2040 if (d_in_lookup(child)) {
2041 struct dentry *res;
2042 res = instantiate(child, task, ptr);
2043 d_lookup_done(child);
2044 if (unlikely(res)) {
2045 dput(child);
2046 child = res;
2047 if (IS_ERR(child))
2048 goto end_instantiate;
2049 }
2050 }
2051 }
2052 inode = d_inode(child);
2053 ino = inode->i_ino;
2054 type = inode->i_mode >> 12;
2055 dput(child);
2056end_instantiate:
2057 return dir_emit(ctx, name, len, ino, type);
2058}
2059
2060/*
2061 * dname_to_vma_addr - maps a dentry name into two unsigned longs
2062 * which represent vma start and end addresses.
2063 */
2064static int dname_to_vma_addr(struct dentry *dentry,
2065 unsigned long *start, unsigned long *end)
2066{
2067 const char *str = dentry->d_name.name;
2068 unsigned long long sval, eval;
2069 unsigned int len;
2070
2071 if (str[0] == '0' && str[1] != '-')
2072 return -EINVAL;
2073 len = _parse_integer(str, 16, &sval);
2074 if (len & KSTRTOX_OVERFLOW)
2075 return -EINVAL;
2076 if (sval != (unsigned long)sval)
2077 return -EINVAL;
2078 str += len;
2079
2080 if (*str != '-')
2081 return -EINVAL;
2082 str++;
2083
2084 if (str[0] == '0' && str[1])
2085 return -EINVAL;
2086 len = _parse_integer(str, 16, &eval);
2087 if (len & KSTRTOX_OVERFLOW)
2088 return -EINVAL;
2089 if (eval != (unsigned long)eval)
2090 return -EINVAL;
2091 str += len;
2092
2093 if (*str != '\0')
2094 return -EINVAL;
2095
2096 *start = sval;
2097 *end = eval;
2098
2099 return 0;
2100}
2101
2102static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
2103{
2104 unsigned long vm_start, vm_end;
2105 bool exact_vma_exists = false;
2106 struct mm_struct *mm = NULL;
2107 struct task_struct *task;
2108 struct inode *inode;
2109 int status = 0;
2110
2111 if (flags & LOOKUP_RCU)
2112 return -ECHILD;
2113
2114 inode = d_inode(dentry);
2115 task = get_proc_task(inode);
2116 if (!task)
2117 goto out_notask;
2118
2119 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
2120 if (IS_ERR_OR_NULL(mm))
2121 goto out;
2122
2123 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2124 status = mmap_read_lock_killable(mm);
2125 if (!status) {
2126 exact_vma_exists = !!find_exact_vma(mm, vm_start,
2127 vm_end);
2128 mmap_read_unlock(mm);
2129 }
2130 }
2131
2132 mmput(mm);
2133
2134 if (exact_vma_exists) {
2135 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
2136
2137 security_task_to_inode(task, inode);
2138 status = 1;
2139 }
2140
2141out:
2142 put_task_struct(task);
2143
2144out_notask:
2145 return status;
2146}
2147
2148static const struct dentry_operations tid_map_files_dentry_operations = {
2149 .d_revalidate = map_files_d_revalidate,
2150 .d_delete = pid_delete_dentry,
2151};
2152
2153static int map_files_get_link(struct dentry *dentry, struct path *path)
2154{
2155 unsigned long vm_start, vm_end;
2156 struct vm_area_struct *vma;
2157 struct task_struct *task;
2158 struct mm_struct *mm;
2159 int rc;
2160
2161 rc = -ENOENT;
2162 task = get_proc_task(d_inode(dentry));
2163 if (!task)
2164 goto out;
2165
2166 mm = get_task_mm(task);
2167 put_task_struct(task);
2168 if (!mm)
2169 goto out;
2170
2171 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2172 if (rc)
2173 goto out_mmput;
2174
2175 rc = mmap_read_lock_killable(mm);
2176 if (rc)
2177 goto out_mmput;
2178
2179 rc = -ENOENT;
2180 vma = find_exact_vma(mm, vm_start, vm_end);
2181 if (vma && vma->vm_file) {
2182 *path = vma->vm_file->f_path;
2183 path_get(path);
2184 rc = 0;
2185 }
2186 mmap_read_unlock(mm);
2187
2188out_mmput:
2189 mmput(mm);
2190out:
2191 return rc;
2192}
2193
2194struct map_files_info {
2195 unsigned long start;
2196 unsigned long end;
2197 fmode_t mode;
2198};
2199
2200/*
2201 * Only allow CAP_SYS_ADMIN and CAP_CHECKPOINT_RESTORE to follow the links, due
2202 * to concerns about how the symlinks may be used to bypass permissions on
2203 * ancestor directories in the path to the file in question.
2204 */
2205static const char *
2206proc_map_files_get_link(struct dentry *dentry,
2207 struct inode *inode,
2208 struct delayed_call *done)
2209{
2210 if (!checkpoint_restore_ns_capable(&init_user_ns))
2211 return ERR_PTR(-EPERM);
2212
2213 return proc_pid_get_link(dentry, inode, done);
2214}
2215
2216/*
2217 * Identical to proc_pid_link_inode_operations except for get_link()
2218 */
2219static const struct inode_operations proc_map_files_link_inode_operations = {
2220 .readlink = proc_pid_readlink,
2221 .get_link = proc_map_files_get_link,
2222 .setattr = proc_setattr,
2223};
2224
2225static struct dentry *
2226proc_map_files_instantiate(struct dentry *dentry,
2227 struct task_struct *task, const void *ptr)
2228{
2229 fmode_t mode = (fmode_t)(unsigned long)ptr;
2230 struct proc_inode *ei;
2231 struct inode *inode;
2232
2233 inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2234 ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2235 ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2236 if (!inode)
2237 return ERR_PTR(-ENOENT);
2238
2239 ei = PROC_I(inode);
2240 ei->op.proc_get_link = map_files_get_link;
2241
2242 inode->i_op = &proc_map_files_link_inode_operations;
2243 inode->i_size = 64;
2244
2245 d_set_d_op(dentry, &tid_map_files_dentry_operations);
2246 return d_splice_alias(inode, dentry);
2247}
2248
2249static struct dentry *proc_map_files_lookup(struct inode *dir,
2250 struct dentry *dentry, unsigned int flags)
2251{
2252 unsigned long vm_start, vm_end;
2253 struct vm_area_struct *vma;
2254 struct task_struct *task;
2255 struct dentry *result;
2256 struct mm_struct *mm;
2257
2258 result = ERR_PTR(-ENOENT);
2259 task = get_proc_task(dir);
2260 if (!task)
2261 goto out;
2262
2263 result = ERR_PTR(-EACCES);
2264 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2265 goto out_put_task;
2266
2267 result = ERR_PTR(-ENOENT);
2268 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2269 goto out_put_task;
2270
2271 mm = get_task_mm(task);
2272 if (!mm)
2273 goto out_put_task;
2274
2275 result = ERR_PTR(-EINTR);
2276 if (mmap_read_lock_killable(mm))
2277 goto out_put_mm;
2278
2279 result = ERR_PTR(-ENOENT);
2280 vma = find_exact_vma(mm, vm_start, vm_end);
2281 if (!vma)
2282 goto out_no_vma;
2283
2284 if (vma->vm_file)
2285 result = proc_map_files_instantiate(dentry, task,
2286 (void *)(unsigned long)vma->vm_file->f_mode);
2287
2288out_no_vma:
2289 mmap_read_unlock(mm);
2290out_put_mm:
2291 mmput(mm);
2292out_put_task:
2293 put_task_struct(task);
2294out:
2295 return result;
2296}
2297
2298static const struct inode_operations proc_map_files_inode_operations = {
2299 .lookup = proc_map_files_lookup,
2300 .permission = proc_fd_permission,
2301 .setattr = proc_setattr,
2302};
2303
2304static int
2305proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2306{
2307 struct vm_area_struct *vma;
2308 struct task_struct *task;
2309 struct mm_struct *mm;
2310 unsigned long nr_files, pos, i;
2311 GENRADIX(struct map_files_info) fa;
2312 struct map_files_info *p;
2313 int ret;
2314
2315 genradix_init(&fa);
2316
2317 ret = -ENOENT;
2318 task = get_proc_task(file_inode(file));
2319 if (!task)
2320 goto out;
2321
2322 ret = -EACCES;
2323 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2324 goto out_put_task;
2325
2326 ret = 0;
2327 if (!dir_emit_dots(file, ctx))
2328 goto out_put_task;
2329
2330 mm = get_task_mm(task);
2331 if (!mm)
2332 goto out_put_task;
2333
2334 ret = mmap_read_lock_killable(mm);
2335 if (ret) {
2336 mmput(mm);
2337 goto out_put_task;
2338 }
2339
2340 nr_files = 0;
2341
2342 /*
2343 * We need two passes here:
2344 *
2345 * 1) Collect vmas of mapped files with mmap_lock taken
2346 * 2) Release mmap_lock and instantiate entries
2347 *
2348 * otherwise we get lockdep complained, since filldir()
2349 * routine might require mmap_lock taken in might_fault().
2350 */
2351
2352 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2353 if (!vma->vm_file)
2354 continue;
2355 if (++pos <= ctx->pos)
2356 continue;
2357
2358 p = genradix_ptr_alloc(&fa, nr_files++, GFP_KERNEL);
2359 if (!p) {
2360 ret = -ENOMEM;
2361 mmap_read_unlock(mm);
2362 mmput(mm);
2363 goto out_put_task;
2364 }
2365
2366 p->start = vma->vm_start;
2367 p->end = vma->vm_end;
2368 p->mode = vma->vm_file->f_mode;
2369 }
2370 mmap_read_unlock(mm);
2371 mmput(mm);
2372
2373 for (i = 0; i < nr_files; i++) {
2374 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2375 unsigned int len;
2376
2377 p = genradix_ptr(&fa, i);
2378 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2379 if (!proc_fill_cache(file, ctx,
2380 buf, len,
2381 proc_map_files_instantiate,
2382 task,
2383 (void *)(unsigned long)p->mode))
2384 break;
2385 ctx->pos++;
2386 }
2387
2388out_put_task:
2389 put_task_struct(task);
2390out:
2391 genradix_free(&fa);
2392 return ret;
2393}
2394
2395static const struct file_operations proc_map_files_operations = {
2396 .read = generic_read_dir,
2397 .iterate_shared = proc_map_files_readdir,
2398 .llseek = generic_file_llseek,
2399};
2400
2401#if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2402struct timers_private {
2403 struct pid *pid;
2404 struct task_struct *task;
2405 struct sighand_struct *sighand;
2406 struct pid_namespace *ns;
2407 unsigned long flags;
2408};
2409
2410static void *timers_start(struct seq_file *m, loff_t *pos)
2411{
2412 struct timers_private *tp = m->private;
2413
2414 tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2415 if (!tp->task)
2416 return ERR_PTR(-ESRCH);
2417
2418 tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2419 if (!tp->sighand)
2420 return ERR_PTR(-ESRCH);
2421
2422 return seq_list_start(&tp->task->signal->posix_timers, *pos);
2423}
2424
2425static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2426{
2427 struct timers_private *tp = m->private;
2428 return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2429}
2430
2431static void timers_stop(struct seq_file *m, void *v)
2432{
2433 struct timers_private *tp = m->private;
2434
2435 if (tp->sighand) {
2436 unlock_task_sighand(tp->task, &tp->flags);
2437 tp->sighand = NULL;
2438 }
2439
2440 if (tp->task) {
2441 put_task_struct(tp->task);
2442 tp->task = NULL;
2443 }
2444}
2445
2446static int show_timer(struct seq_file *m, void *v)
2447{
2448 struct k_itimer *timer;
2449 struct timers_private *tp = m->private;
2450 int notify;
2451 static const char * const nstr[] = {
2452 [SIGEV_SIGNAL] = "signal",
2453 [SIGEV_NONE] = "none",
2454 [SIGEV_THREAD] = "thread",
2455 };
2456
2457 timer = list_entry((struct list_head *)v, struct k_itimer, list);
2458 notify = timer->it_sigev_notify;
2459
2460 seq_printf(m, "ID: %d\n", timer->it_id);
2461 seq_printf(m, "signal: %d/%px\n",
2462 timer->sigq->info.si_signo,
2463 timer->sigq->info.si_value.sival_ptr);
2464 seq_printf(m, "notify: %s/%s.%d\n",
2465 nstr[notify & ~SIGEV_THREAD_ID],
2466 (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2467 pid_nr_ns(timer->it_pid, tp->ns));
2468 seq_printf(m, "ClockID: %d\n", timer->it_clock);
2469
2470 return 0;
2471}
2472
2473static const struct seq_operations proc_timers_seq_ops = {
2474 .start = timers_start,
2475 .next = timers_next,
2476 .stop = timers_stop,
2477 .show = show_timer,
2478};
2479
2480static int proc_timers_open(struct inode *inode, struct file *file)
2481{
2482 struct timers_private *tp;
2483
2484 tp = __seq_open_private(file, &proc_timers_seq_ops,
2485 sizeof(struct timers_private));
2486 if (!tp)
2487 return -ENOMEM;
2488
2489 tp->pid = proc_pid(inode);
2490 tp->ns = proc_pid_ns(inode->i_sb);
2491 return 0;
2492}
2493
2494static const struct file_operations proc_timers_operations = {
2495 .open = proc_timers_open,
2496 .read = seq_read,
2497 .llseek = seq_lseek,
2498 .release = seq_release_private,
2499};
2500#endif
2501
2502static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2503 size_t count, loff_t *offset)
2504{
2505 struct inode *inode = file_inode(file);
2506 struct task_struct *p;
2507 u64 slack_ns;
2508 int err;
2509
2510 err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2511 if (err < 0)
2512 return err;
2513
2514 p = get_proc_task(inode);
2515 if (!p)
2516 return -ESRCH;
2517
2518 if (p != current) {
2519 rcu_read_lock();
2520 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2521 rcu_read_unlock();
2522 count = -EPERM;
2523 goto out;
2524 }
2525 rcu_read_unlock();
2526
2527 err = security_task_setscheduler(p);
2528 if (err) {
2529 count = err;
2530 goto out;
2531 }
2532 }
2533
2534 task_lock(p);
2535 if (slack_ns == 0)
2536 p->timer_slack_ns = p->default_timer_slack_ns;
2537 else
2538 p->timer_slack_ns = slack_ns;
2539 task_unlock(p);
2540
2541out:
2542 put_task_struct(p);
2543
2544 return count;
2545}
2546
2547static int timerslack_ns_show(struct seq_file *m, void *v)
2548{
2549 struct inode *inode = m->private;
2550 struct task_struct *p;
2551 int err = 0;
2552
2553 p = get_proc_task(inode);
2554 if (!p)
2555 return -ESRCH;
2556
2557 if (p != current) {
2558 rcu_read_lock();
2559 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2560 rcu_read_unlock();
2561 err = -EPERM;
2562 goto out;
2563 }
2564 rcu_read_unlock();
2565
2566 err = security_task_getscheduler(p);
2567 if (err)
2568 goto out;
2569 }
2570
2571 task_lock(p);
2572 seq_printf(m, "%llu\n", p->timer_slack_ns);
2573 task_unlock(p);
2574
2575out:
2576 put_task_struct(p);
2577
2578 return err;
2579}
2580
2581static int timerslack_ns_open(struct inode *inode, struct file *filp)
2582{
2583 return single_open(filp, timerslack_ns_show, inode);
2584}
2585
2586static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2587 .open = timerslack_ns_open,
2588 .read = seq_read,
2589 .write = timerslack_ns_write,
2590 .llseek = seq_lseek,
2591 .release = single_release,
2592};
2593
2594static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2595 struct task_struct *task, const void *ptr)
2596{
2597 const struct pid_entry *p = ptr;
2598 struct inode *inode;
2599 struct proc_inode *ei;
2600
2601 inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2602 if (!inode)
2603 return ERR_PTR(-ENOENT);
2604
2605 ei = PROC_I(inode);
2606 if (S_ISDIR(inode->i_mode))
2607 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2608 if (p->iop)
2609 inode->i_op = p->iop;
2610 if (p->fop)
2611 inode->i_fop = p->fop;
2612 ei->op = p->op;
2613 pid_update_inode(task, inode);
2614 d_set_d_op(dentry, &pid_dentry_operations);
2615 return d_splice_alias(inode, dentry);
2616}
2617
2618static struct dentry *proc_pident_lookup(struct inode *dir,
2619 struct dentry *dentry,
2620 const struct pid_entry *p,
2621 const struct pid_entry *end)
2622{
2623 struct task_struct *task = get_proc_task(dir);
2624 struct dentry *res = ERR_PTR(-ENOENT);
2625
2626 if (!task)
2627 goto out_no_task;
2628
2629 /*
2630 * Yes, it does not scale. And it should not. Don't add
2631 * new entries into /proc/<tgid>/ without very good reasons.
2632 */
2633 for (; p < end; p++) {
2634 if (p->len != dentry->d_name.len)
2635 continue;
2636 if (!memcmp(dentry->d_name.name, p->name, p->len)) {
2637 res = proc_pident_instantiate(dentry, task, p);
2638 break;
2639 }
2640 }
2641 put_task_struct(task);
2642out_no_task:
2643 return res;
2644}
2645
2646static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2647 const struct pid_entry *ents, unsigned int nents)
2648{
2649 struct task_struct *task = get_proc_task(file_inode(file));
2650 const struct pid_entry *p;
2651
2652 if (!task)
2653 return -ENOENT;
2654
2655 if (!dir_emit_dots(file, ctx))
2656 goto out;
2657
2658 if (ctx->pos >= nents + 2)
2659 goto out;
2660
2661 for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2662 if (!proc_fill_cache(file, ctx, p->name, p->len,
2663 proc_pident_instantiate, task, p))
2664 break;
2665 ctx->pos++;
2666 }
2667out:
2668 put_task_struct(task);
2669 return 0;
2670}
2671
2672#ifdef CONFIG_SECURITY
2673static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2674 size_t count, loff_t *ppos)
2675{
2676 struct inode * inode = file_inode(file);
2677 char *p = NULL;
2678 ssize_t length;
2679 struct task_struct *task = get_proc_task(inode);
2680
2681 if (!task)
2682 return -ESRCH;
2683
2684 length = security_getprocattr(task, PROC_I(inode)->op.lsm,
2685 (char*)file->f_path.dentry->d_name.name,
2686 &p);
2687 put_task_struct(task);
2688 if (length > 0)
2689 length = simple_read_from_buffer(buf, count, ppos, p, length);
2690 kfree(p);
2691 return length;
2692}
2693
2694static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2695 size_t count, loff_t *ppos)
2696{
2697 struct inode * inode = file_inode(file);
2698 struct task_struct *task;
2699 void *page;
2700 int rv;
2701
2702 rcu_read_lock();
2703 task = pid_task(proc_pid(inode), PIDTYPE_PID);
2704 if (!task) {
2705 rcu_read_unlock();
2706 return -ESRCH;
2707 }
2708 /* A task may only write its own attributes. */
2709 if (current != task) {
2710 rcu_read_unlock();
2711 return -EACCES;
2712 }
2713 /* Prevent changes to overridden credentials. */
2714 if (current_cred() != current_real_cred()) {
2715 rcu_read_unlock();
2716 return -EBUSY;
2717 }
2718 rcu_read_unlock();
2719
2720 if (count > PAGE_SIZE)
2721 count = PAGE_SIZE;
2722
2723 /* No partial writes. */
2724 if (*ppos != 0)
2725 return -EINVAL;
2726
2727 page = memdup_user(buf, count);
2728 if (IS_ERR(page)) {
2729 rv = PTR_ERR(page);
2730 goto out;
2731 }
2732
2733 /* Guard against adverse ptrace interaction */
2734 rv = mutex_lock_interruptible(¤t->signal->cred_guard_mutex);
2735 if (rv < 0)
2736 goto out_free;
2737
2738 rv = security_setprocattr(PROC_I(inode)->op.lsm,
2739 file->f_path.dentry->d_name.name, page,
2740 count);
2741 mutex_unlock(¤t->signal->cred_guard_mutex);
2742out_free:
2743 kfree(page);
2744out:
2745 return rv;
2746}
2747
2748static const struct file_operations proc_pid_attr_operations = {
2749 .read = proc_pid_attr_read,
2750 .write = proc_pid_attr_write,
2751 .llseek = generic_file_llseek,
2752};
2753
2754#define LSM_DIR_OPS(LSM) \
2755static int proc_##LSM##_attr_dir_iterate(struct file *filp, \
2756 struct dir_context *ctx) \
2757{ \
2758 return proc_pident_readdir(filp, ctx, \
2759 LSM##_attr_dir_stuff, \
2760 ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2761} \
2762\
2763static const struct file_operations proc_##LSM##_attr_dir_ops = { \
2764 .read = generic_read_dir, \
2765 .iterate = proc_##LSM##_attr_dir_iterate, \
2766 .llseek = default_llseek, \
2767}; \
2768\
2769static struct dentry *proc_##LSM##_attr_dir_lookup(struct inode *dir, \
2770 struct dentry *dentry, unsigned int flags) \
2771{ \
2772 return proc_pident_lookup(dir, dentry, \
2773 LSM##_attr_dir_stuff, \
2774 LSM##_attr_dir_stuff + ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2775} \
2776\
2777static const struct inode_operations proc_##LSM##_attr_dir_inode_ops = { \
2778 .lookup = proc_##LSM##_attr_dir_lookup, \
2779 .getattr = pid_getattr, \
2780 .setattr = proc_setattr, \
2781}
2782
2783#ifdef CONFIG_SECURITY_SMACK
2784static const struct pid_entry smack_attr_dir_stuff[] = {
2785 ATTR("smack", "current", 0666),
2786};
2787LSM_DIR_OPS(smack);
2788#endif
2789
2790#ifdef CONFIG_SECURITY_APPARMOR
2791static const struct pid_entry apparmor_attr_dir_stuff[] = {
2792 ATTR("apparmor", "current", 0666),
2793 ATTR("apparmor", "prev", 0444),
2794 ATTR("apparmor", "exec", 0666),
2795};
2796LSM_DIR_OPS(apparmor);
2797#endif
2798
2799static const struct pid_entry attr_dir_stuff[] = {
2800 ATTR(NULL, "current", 0666),
2801 ATTR(NULL, "prev", 0444),
2802 ATTR(NULL, "exec", 0666),
2803 ATTR(NULL, "fscreate", 0666),
2804 ATTR(NULL, "keycreate", 0666),
2805 ATTR(NULL, "sockcreate", 0666),
2806#ifdef CONFIG_SECURITY_SMACK
2807 DIR("smack", 0555,
2808 proc_smack_attr_dir_inode_ops, proc_smack_attr_dir_ops),
2809#endif
2810#ifdef CONFIG_SECURITY_APPARMOR
2811 DIR("apparmor", 0555,
2812 proc_apparmor_attr_dir_inode_ops, proc_apparmor_attr_dir_ops),
2813#endif
2814};
2815
2816static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2817{
2818 return proc_pident_readdir(file, ctx,
2819 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2820}
2821
2822static const struct file_operations proc_attr_dir_operations = {
2823 .read = generic_read_dir,
2824 .iterate_shared = proc_attr_dir_readdir,
2825 .llseek = generic_file_llseek,
2826};
2827
2828static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2829 struct dentry *dentry, unsigned int flags)
2830{
2831 return proc_pident_lookup(dir, dentry,
2832 attr_dir_stuff,
2833 attr_dir_stuff + ARRAY_SIZE(attr_dir_stuff));
2834}
2835
2836static const struct inode_operations proc_attr_dir_inode_operations = {
2837 .lookup = proc_attr_dir_lookup,
2838 .getattr = pid_getattr,
2839 .setattr = proc_setattr,
2840};
2841
2842#endif
2843
2844#ifdef CONFIG_ELF_CORE
2845static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2846 size_t count, loff_t *ppos)
2847{
2848 struct task_struct *task = get_proc_task(file_inode(file));
2849 struct mm_struct *mm;
2850 char buffer[PROC_NUMBUF];
2851 size_t len;
2852 int ret;
2853
2854 if (!task)
2855 return -ESRCH;
2856
2857 ret = 0;
2858 mm = get_task_mm(task);
2859 if (mm) {
2860 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2861 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2862 MMF_DUMP_FILTER_SHIFT));
2863 mmput(mm);
2864 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2865 }
2866
2867 put_task_struct(task);
2868
2869 return ret;
2870}
2871
2872static ssize_t proc_coredump_filter_write(struct file *file,
2873 const char __user *buf,
2874 size_t count,
2875 loff_t *ppos)
2876{
2877 struct task_struct *task;
2878 struct mm_struct *mm;
2879 unsigned int val;
2880 int ret;
2881 int i;
2882 unsigned long mask;
2883
2884 ret = kstrtouint_from_user(buf, count, 0, &val);
2885 if (ret < 0)
2886 return ret;
2887
2888 ret = -ESRCH;
2889 task = get_proc_task(file_inode(file));
2890 if (!task)
2891 goto out_no_task;
2892
2893 mm = get_task_mm(task);
2894 if (!mm)
2895 goto out_no_mm;
2896 ret = 0;
2897
2898 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2899 if (val & mask)
2900 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2901 else
2902 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2903 }
2904
2905 mmput(mm);
2906 out_no_mm:
2907 put_task_struct(task);
2908 out_no_task:
2909 if (ret < 0)
2910 return ret;
2911 return count;
2912}
2913
2914static const struct file_operations proc_coredump_filter_operations = {
2915 .read = proc_coredump_filter_read,
2916 .write = proc_coredump_filter_write,
2917 .llseek = generic_file_llseek,
2918};
2919#endif
2920
2921#ifdef CONFIG_TASK_IO_ACCOUNTING
2922static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2923{
2924 struct task_io_accounting acct = task->ioac;
2925 unsigned long flags;
2926 int result;
2927
2928 result = mutex_lock_killable(&task->signal->exec_update_mutex);
2929 if (result)
2930 return result;
2931
2932 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2933 result = -EACCES;
2934 goto out_unlock;
2935 }
2936
2937 if (whole && lock_task_sighand(task, &flags)) {
2938 struct task_struct *t = task;
2939
2940 task_io_accounting_add(&acct, &task->signal->ioac);
2941 while_each_thread(task, t)
2942 task_io_accounting_add(&acct, &t->ioac);
2943
2944 unlock_task_sighand(task, &flags);
2945 }
2946 seq_printf(m,
2947 "rchar: %llu\n"
2948 "wchar: %llu\n"
2949 "syscr: %llu\n"
2950 "syscw: %llu\n"
2951 "read_bytes: %llu\n"
2952 "write_bytes: %llu\n"
2953 "cancelled_write_bytes: %llu\n",
2954 (unsigned long long)acct.rchar,
2955 (unsigned long long)acct.wchar,
2956 (unsigned long long)acct.syscr,
2957 (unsigned long long)acct.syscw,
2958 (unsigned long long)acct.read_bytes,
2959 (unsigned long long)acct.write_bytes,
2960 (unsigned long long)acct.cancelled_write_bytes);
2961 result = 0;
2962
2963out_unlock:
2964 mutex_unlock(&task->signal->exec_update_mutex);
2965 return result;
2966}
2967
2968static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2969 struct pid *pid, struct task_struct *task)
2970{
2971 return do_io_accounting(task, m, 0);
2972}
2973
2974static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2975 struct pid *pid, struct task_struct *task)
2976{
2977 return do_io_accounting(task, m, 1);
2978}
2979#endif /* CONFIG_TASK_IO_ACCOUNTING */
2980
2981#ifdef CONFIG_USER_NS
2982static int proc_id_map_open(struct inode *inode, struct file *file,
2983 const struct seq_operations *seq_ops)
2984{
2985 struct user_namespace *ns = NULL;
2986 struct task_struct *task;
2987 struct seq_file *seq;
2988 int ret = -EINVAL;
2989
2990 task = get_proc_task(inode);
2991 if (task) {
2992 rcu_read_lock();
2993 ns = get_user_ns(task_cred_xxx(task, user_ns));
2994 rcu_read_unlock();
2995 put_task_struct(task);
2996 }
2997 if (!ns)
2998 goto err;
2999
3000 ret = seq_open(file, seq_ops);
3001 if (ret)
3002 goto err_put_ns;
3003
3004 seq = file->private_data;
3005 seq->private = ns;
3006
3007 return 0;
3008err_put_ns:
3009 put_user_ns(ns);
3010err:
3011 return ret;
3012}
3013
3014static int proc_id_map_release(struct inode *inode, struct file *file)
3015{
3016 struct seq_file *seq = file->private_data;
3017 struct user_namespace *ns = seq->private;
3018 put_user_ns(ns);
3019 return seq_release(inode, file);
3020}
3021
3022static int proc_uid_map_open(struct inode *inode, struct file *file)
3023{
3024 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
3025}
3026
3027static int proc_gid_map_open(struct inode *inode, struct file *file)
3028{
3029 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
3030}
3031
3032static int proc_projid_map_open(struct inode *inode, struct file *file)
3033{
3034 return proc_id_map_open(inode, file, &proc_projid_seq_operations);
3035}
3036
3037static const struct file_operations proc_uid_map_operations = {
3038 .open = proc_uid_map_open,
3039 .write = proc_uid_map_write,
3040 .read = seq_read,
3041 .llseek = seq_lseek,
3042 .release = proc_id_map_release,
3043};
3044
3045static const struct file_operations proc_gid_map_operations = {
3046 .open = proc_gid_map_open,
3047 .write = proc_gid_map_write,
3048 .read = seq_read,
3049 .llseek = seq_lseek,
3050 .release = proc_id_map_release,
3051};
3052
3053static const struct file_operations proc_projid_map_operations = {
3054 .open = proc_projid_map_open,
3055 .write = proc_projid_map_write,
3056 .read = seq_read,
3057 .llseek = seq_lseek,
3058 .release = proc_id_map_release,
3059};
3060
3061static int proc_setgroups_open(struct inode *inode, struct file *file)
3062{
3063 struct user_namespace *ns = NULL;
3064 struct task_struct *task;
3065 int ret;
3066
3067 ret = -ESRCH;
3068 task = get_proc_task(inode);
3069 if (task) {
3070 rcu_read_lock();
3071 ns = get_user_ns(task_cred_xxx(task, user_ns));
3072 rcu_read_unlock();
3073 put_task_struct(task);
3074 }
3075 if (!ns)
3076 goto err;
3077
3078 if (file->f_mode & FMODE_WRITE) {
3079 ret = -EACCES;
3080 if (!ns_capable(ns, CAP_SYS_ADMIN))
3081 goto err_put_ns;
3082 }
3083
3084 ret = single_open(file, &proc_setgroups_show, ns);
3085 if (ret)
3086 goto err_put_ns;
3087
3088 return 0;
3089err_put_ns:
3090 put_user_ns(ns);
3091err:
3092 return ret;
3093}
3094
3095static int proc_setgroups_release(struct inode *inode, struct file *file)
3096{
3097 struct seq_file *seq = file->private_data;
3098 struct user_namespace *ns = seq->private;
3099 int ret = single_release(inode, file);
3100 put_user_ns(ns);
3101 return ret;
3102}
3103
3104static const struct file_operations proc_setgroups_operations = {
3105 .open = proc_setgroups_open,
3106 .write = proc_setgroups_write,
3107 .read = seq_read,
3108 .llseek = seq_lseek,
3109 .release = proc_setgroups_release,
3110};
3111#endif /* CONFIG_USER_NS */
3112
3113static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
3114 struct pid *pid, struct task_struct *task)
3115{
3116 int err = lock_trace(task);
3117 if (!err) {
3118 seq_printf(m, "%08x\n", task->personality);
3119 unlock_trace(task);
3120 }
3121 return err;
3122}
3123
3124#ifdef CONFIG_LIVEPATCH
3125static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
3126 struct pid *pid, struct task_struct *task)
3127{
3128 seq_printf(m, "%d\n", task->patch_state);
3129 return 0;
3130}
3131#endif /* CONFIG_LIVEPATCH */
3132
3133#ifdef CONFIG_STACKLEAK_METRICS
3134static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns,
3135 struct pid *pid, struct task_struct *task)
3136{
3137 unsigned long prev_depth = THREAD_SIZE -
3138 (task->prev_lowest_stack & (THREAD_SIZE - 1));
3139 unsigned long depth = THREAD_SIZE -
3140 (task->lowest_stack & (THREAD_SIZE - 1));
3141
3142 seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n",
3143 prev_depth, depth);
3144 return 0;
3145}
3146#endif /* CONFIG_STACKLEAK_METRICS */
3147
3148/*
3149 * Thread groups
3150 */
3151static const struct file_operations proc_task_operations;
3152static const struct inode_operations proc_task_inode_operations;
3153
3154static const struct pid_entry tgid_base_stuff[] = {
3155 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3156 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3157 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3158 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3159 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3160#ifdef CONFIG_NET
3161 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3162#endif
3163 REG("environ", S_IRUSR, proc_environ_operations),
3164 REG("auxv", S_IRUSR, proc_auxv_operations),
3165 ONE("status", S_IRUGO, proc_pid_status),
3166 ONE("personality", S_IRUSR, proc_pid_personality),
3167 ONE("limits", S_IRUGO, proc_pid_limits),
3168#ifdef CONFIG_SCHED_DEBUG
3169 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3170#endif
3171#ifdef CONFIG_SCHED_AUTOGROUP
3172 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3173#endif
3174#ifdef CONFIG_TIME_NS
3175 REG("timens_offsets", S_IRUGO|S_IWUSR, proc_timens_offsets_operations),
3176#endif
3177 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3178#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3179 ONE("syscall", S_IRUSR, proc_pid_syscall),
3180#endif
3181 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3182 ONE("stat", S_IRUGO, proc_tgid_stat),
3183 ONE("statm", S_IRUGO, proc_pid_statm),
3184 REG("maps", S_IRUGO, proc_pid_maps_operations),
3185#ifdef CONFIG_NUMA
3186 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3187#endif
3188 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3189 LNK("cwd", proc_cwd_link),
3190 LNK("root", proc_root_link),
3191 LNK("exe", proc_exe_link),
3192 REG("mounts", S_IRUGO, proc_mounts_operations),
3193 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3194 REG("mountstats", S_IRUSR, proc_mountstats_operations),
3195#ifdef CONFIG_PROC_PAGE_MONITOR
3196 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3197 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3198 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3199 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3200#endif
3201#ifdef CONFIG_SECURITY
3202 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3203#endif
3204#ifdef CONFIG_KALLSYMS
3205 ONE("wchan", S_IRUGO, proc_pid_wchan),
3206#endif
3207#ifdef CONFIG_STACKTRACE
3208 ONE("stack", S_IRUSR, proc_pid_stack),
3209#endif
3210#ifdef CONFIG_SCHED_INFO
3211 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3212#endif
3213#ifdef CONFIG_LATENCYTOP
3214 REG("latency", S_IRUGO, proc_lstats_operations),
3215#endif
3216#ifdef CONFIG_PROC_PID_CPUSET
3217 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3218#endif
3219#ifdef CONFIG_CGROUPS
3220 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3221#endif
3222#ifdef CONFIG_PROC_CPU_RESCTRL
3223 ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3224#endif
3225 ONE("oom_score", S_IRUGO, proc_oom_score),
3226 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3227 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3228#ifdef CONFIG_AUDIT
3229 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3230 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3231#endif
3232#ifdef CONFIG_FAULT_INJECTION
3233 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3234 REG("fail-nth", 0644, proc_fail_nth_operations),
3235#endif
3236#ifdef CONFIG_ELF_CORE
3237 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3238#endif
3239#ifdef CONFIG_TASK_IO_ACCOUNTING
3240 ONE("io", S_IRUSR, proc_tgid_io_accounting),
3241#endif
3242#ifdef CONFIG_USER_NS
3243 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3244 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3245 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3246 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3247#endif
3248#if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3249 REG("timers", S_IRUGO, proc_timers_operations),
3250#endif
3251 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3252#ifdef CONFIG_LIVEPATCH
3253 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3254#endif
3255#ifdef CONFIG_STACKLEAK_METRICS
3256 ONE("stack_depth", S_IRUGO, proc_stack_depth),
3257#endif
3258#ifdef CONFIG_PROC_PID_ARCH_STATUS
3259 ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3260#endif
3261};
3262
3263static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3264{
3265 return proc_pident_readdir(file, ctx,
3266 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3267}
3268
3269static const struct file_operations proc_tgid_base_operations = {
3270 .read = generic_read_dir,
3271 .iterate_shared = proc_tgid_base_readdir,
3272 .llseek = generic_file_llseek,
3273};
3274
3275struct pid *tgid_pidfd_to_pid(const struct file *file)
3276{
3277 if (file->f_op != &proc_tgid_base_operations)
3278 return ERR_PTR(-EBADF);
3279
3280 return proc_pid(file_inode(file));
3281}
3282
3283static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3284{
3285 return proc_pident_lookup(dir, dentry,
3286 tgid_base_stuff,
3287 tgid_base_stuff + ARRAY_SIZE(tgid_base_stuff));
3288}
3289
3290static const struct inode_operations proc_tgid_base_inode_operations = {
3291 .lookup = proc_tgid_base_lookup,
3292 .getattr = pid_getattr,
3293 .setattr = proc_setattr,
3294 .permission = proc_pid_permission,
3295};
3296
3297/**
3298 * proc_flush_pid - Remove dcache entries for @pid from the /proc dcache.
3299 * @pid: pid that should be flushed.
3300 *
3301 * This function walks a list of inodes (that belong to any proc
3302 * filesystem) that are attached to the pid and flushes them from
3303 * the dentry cache.
3304 *
3305 * It is safe and reasonable to cache /proc entries for a task until
3306 * that task exits. After that they just clog up the dcache with
3307 * useless entries, possibly causing useful dcache entries to be
3308 * flushed instead. This routine is provided to flush those useless
3309 * dcache entries when a process is reaped.
3310 *
3311 * NOTE: This routine is just an optimization so it does not guarantee
3312 * that no dcache entries will exist after a process is reaped
3313 * it just makes it very unlikely that any will persist.
3314 */
3315
3316void proc_flush_pid(struct pid *pid)
3317{
3318 proc_invalidate_siblings_dcache(&pid->inodes, &pid->lock);
3319}
3320
3321static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3322 struct task_struct *task, const void *ptr)
3323{
3324 struct inode *inode;
3325
3326 inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3327 if (!inode)
3328 return ERR_PTR(-ENOENT);
3329
3330 inode->i_op = &proc_tgid_base_inode_operations;
3331 inode->i_fop = &proc_tgid_base_operations;
3332 inode->i_flags|=S_IMMUTABLE;
3333
3334 set_nlink(inode, nlink_tgid);
3335 pid_update_inode(task, inode);
3336
3337 d_set_d_op(dentry, &pid_dentry_operations);
3338 return d_splice_alias(inode, dentry);
3339}
3340
3341struct dentry *proc_pid_lookup(struct dentry *dentry, unsigned int flags)
3342{
3343 struct task_struct *task;
3344 unsigned tgid;
3345 struct proc_fs_info *fs_info;
3346 struct pid_namespace *ns;
3347 struct dentry *result = ERR_PTR(-ENOENT);
3348
3349 tgid = name_to_int(&dentry->d_name);
3350 if (tgid == ~0U)
3351 goto out;
3352
3353 fs_info = proc_sb_info(dentry->d_sb);
3354 ns = fs_info->pid_ns;
3355 rcu_read_lock();
3356 task = find_task_by_pid_ns(tgid, ns);
3357 if (task)
3358 get_task_struct(task);
3359 rcu_read_unlock();
3360 if (!task)
3361 goto out;
3362
3363 /* Limit procfs to only ptraceable tasks */
3364 if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE) {
3365 if (!has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS))
3366 goto out_put_task;
3367 }
3368
3369 result = proc_pid_instantiate(dentry, task, NULL);
3370out_put_task:
3371 put_task_struct(task);
3372out:
3373 return result;
3374}
3375
3376/*
3377 * Find the first task with tgid >= tgid
3378 *
3379 */
3380struct tgid_iter {
3381 unsigned int tgid;
3382 struct task_struct *task;
3383};
3384static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3385{
3386 struct pid *pid;
3387
3388 if (iter.task)
3389 put_task_struct(iter.task);
3390 rcu_read_lock();
3391retry:
3392 iter.task = NULL;
3393 pid = find_ge_pid(iter.tgid, ns);
3394 if (pid) {
3395 iter.tgid = pid_nr_ns(pid, ns);
3396 iter.task = pid_task(pid, PIDTYPE_TGID);
3397 if (!iter.task) {
3398 iter.tgid += 1;
3399 goto retry;
3400 }
3401 get_task_struct(iter.task);
3402 }
3403 rcu_read_unlock();
3404 return iter;
3405}
3406
3407#define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3408
3409/* for the /proc/ directory itself, after non-process stuff has been done */
3410int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3411{
3412 struct tgid_iter iter;
3413 struct proc_fs_info *fs_info = proc_sb_info(file_inode(file)->i_sb);
3414 struct pid_namespace *ns = proc_pid_ns(file_inode(file)->i_sb);
3415 loff_t pos = ctx->pos;
3416
3417 if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3418 return 0;
3419
3420 if (pos == TGID_OFFSET - 2) {
3421 struct inode *inode = d_inode(fs_info->proc_self);
3422 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3423 return 0;
3424 ctx->pos = pos = pos + 1;
3425 }
3426 if (pos == TGID_OFFSET - 1) {
3427 struct inode *inode = d_inode(fs_info->proc_thread_self);
3428 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3429 return 0;
3430 ctx->pos = pos = pos + 1;
3431 }
3432 iter.tgid = pos - TGID_OFFSET;
3433 iter.task = NULL;
3434 for (iter = next_tgid(ns, iter);
3435 iter.task;
3436 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3437 char name[10 + 1];
3438 unsigned int len;
3439
3440 cond_resched();
3441 if (!has_pid_permissions(fs_info, iter.task, HIDEPID_INVISIBLE))
3442 continue;
3443
3444 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3445 ctx->pos = iter.tgid + TGID_OFFSET;
3446 if (!proc_fill_cache(file, ctx, name, len,
3447 proc_pid_instantiate, iter.task, NULL)) {
3448 put_task_struct(iter.task);
3449 return 0;
3450 }
3451 }
3452 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3453 return 0;
3454}
3455
3456/*
3457 * proc_tid_comm_permission is a special permission function exclusively
3458 * used for the node /proc/<pid>/task/<tid>/comm.
3459 * It bypasses generic permission checks in the case where a task of the same
3460 * task group attempts to access the node.
3461 * The rationale behind this is that glibc and bionic access this node for
3462 * cross thread naming (pthread_set/getname_np(!self)). However, if
3463 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3464 * which locks out the cross thread naming implementation.
3465 * This function makes sure that the node is always accessible for members of
3466 * same thread group.
3467 */
3468static int proc_tid_comm_permission(struct inode *inode, int mask)
3469{
3470 bool is_same_tgroup;
3471 struct task_struct *task;
3472
3473 task = get_proc_task(inode);
3474 if (!task)
3475 return -ESRCH;
3476 is_same_tgroup = same_thread_group(current, task);
3477 put_task_struct(task);
3478
3479 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3480 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3481 * read or written by the members of the corresponding
3482 * thread group.
3483 */
3484 return 0;
3485 }
3486
3487 return generic_permission(inode, mask);
3488}
3489
3490static const struct inode_operations proc_tid_comm_inode_operations = {
3491 .permission = proc_tid_comm_permission,
3492};
3493
3494/*
3495 * Tasks
3496 */
3497static const struct pid_entry tid_base_stuff[] = {
3498 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3499 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3500 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3501#ifdef CONFIG_NET
3502 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3503#endif
3504 REG("environ", S_IRUSR, proc_environ_operations),
3505 REG("auxv", S_IRUSR, proc_auxv_operations),
3506 ONE("status", S_IRUGO, proc_pid_status),
3507 ONE("personality", S_IRUSR, proc_pid_personality),
3508 ONE("limits", S_IRUGO, proc_pid_limits),
3509#ifdef CONFIG_SCHED_DEBUG
3510 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3511#endif
3512 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR,
3513 &proc_tid_comm_inode_operations,
3514 &proc_pid_set_comm_operations, {}),
3515#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3516 ONE("syscall", S_IRUSR, proc_pid_syscall),
3517#endif
3518 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops),
3519 ONE("stat", S_IRUGO, proc_tid_stat),
3520 ONE("statm", S_IRUGO, proc_pid_statm),
3521 REG("maps", S_IRUGO, proc_pid_maps_operations),
3522#ifdef CONFIG_PROC_CHILDREN
3523 REG("children", S_IRUGO, proc_tid_children_operations),
3524#endif
3525#ifdef CONFIG_NUMA
3526 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3527#endif
3528 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3529 LNK("cwd", proc_cwd_link),
3530 LNK("root", proc_root_link),
3531 LNK("exe", proc_exe_link),
3532 REG("mounts", S_IRUGO, proc_mounts_operations),
3533 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3534#ifdef CONFIG_PROC_PAGE_MONITOR
3535 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3536 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3537 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3538 REG("pagemap", S_IRUSR, proc_pagemap_operations),
3539#endif
3540#ifdef CONFIG_SECURITY
3541 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3542#endif
3543#ifdef CONFIG_KALLSYMS
3544 ONE("wchan", S_IRUGO, proc_pid_wchan),
3545#endif
3546#ifdef CONFIG_STACKTRACE
3547 ONE("stack", S_IRUSR, proc_pid_stack),
3548#endif
3549#ifdef CONFIG_SCHED_INFO
3550 ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3551#endif
3552#ifdef CONFIG_LATENCYTOP
3553 REG("latency", S_IRUGO, proc_lstats_operations),
3554#endif
3555#ifdef CONFIG_PROC_PID_CPUSET
3556 ONE("cpuset", S_IRUGO, proc_cpuset_show),
3557#endif
3558#ifdef CONFIG_CGROUPS
3559 ONE("cgroup", S_IRUGO, proc_cgroup_show),
3560#endif
3561#ifdef CONFIG_PROC_CPU_RESCTRL
3562 ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3563#endif
3564 ONE("oom_score", S_IRUGO, proc_oom_score),
3565 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3566 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3567#ifdef CONFIG_AUDIT
3568 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3569 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3570#endif
3571#ifdef CONFIG_FAULT_INJECTION
3572 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3573 REG("fail-nth", 0644, proc_fail_nth_operations),
3574#endif
3575#ifdef CONFIG_TASK_IO_ACCOUNTING
3576 ONE("io", S_IRUSR, proc_tid_io_accounting),
3577#endif
3578#ifdef CONFIG_USER_NS
3579 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3580 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3581 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3582 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations),
3583#endif
3584#ifdef CONFIG_LIVEPATCH
3585 ONE("patch_state", S_IRUSR, proc_pid_patch_state),
3586#endif
3587#ifdef CONFIG_PROC_PID_ARCH_STATUS
3588 ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3589#endif
3590};
3591
3592static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3593{
3594 return proc_pident_readdir(file, ctx,
3595 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3596}
3597
3598static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3599{
3600 return proc_pident_lookup(dir, dentry,
3601 tid_base_stuff,
3602 tid_base_stuff + ARRAY_SIZE(tid_base_stuff));
3603}
3604
3605static const struct file_operations proc_tid_base_operations = {
3606 .read = generic_read_dir,
3607 .iterate_shared = proc_tid_base_readdir,
3608 .llseek = generic_file_llseek,
3609};
3610
3611static const struct inode_operations proc_tid_base_inode_operations = {
3612 .lookup = proc_tid_base_lookup,
3613 .getattr = pid_getattr,
3614 .setattr = proc_setattr,
3615};
3616
3617static struct dentry *proc_task_instantiate(struct dentry *dentry,
3618 struct task_struct *task, const void *ptr)
3619{
3620 struct inode *inode;
3621 inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3622 if (!inode)
3623 return ERR_PTR(-ENOENT);
3624
3625 inode->i_op = &proc_tid_base_inode_operations;
3626 inode->i_fop = &proc_tid_base_operations;
3627 inode->i_flags |= S_IMMUTABLE;
3628
3629 set_nlink(inode, nlink_tid);
3630 pid_update_inode(task, inode);
3631
3632 d_set_d_op(dentry, &pid_dentry_operations);
3633 return d_splice_alias(inode, dentry);
3634}
3635
3636static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3637{
3638 struct task_struct *task;
3639 struct task_struct *leader = get_proc_task(dir);
3640 unsigned tid;
3641 struct proc_fs_info *fs_info;
3642 struct pid_namespace *ns;
3643 struct dentry *result = ERR_PTR(-ENOENT);
3644
3645 if (!leader)
3646 goto out_no_task;
3647
3648 tid = name_to_int(&dentry->d_name);
3649 if (tid == ~0U)
3650 goto out;
3651
3652 fs_info = proc_sb_info(dentry->d_sb);
3653 ns = fs_info->pid_ns;
3654 rcu_read_lock();
3655 task = find_task_by_pid_ns(tid, ns);
3656 if (task)
3657 get_task_struct(task);
3658 rcu_read_unlock();
3659 if (!task)
3660 goto out;
3661 if (!same_thread_group(leader, task))
3662 goto out_drop_task;
3663
3664 result = proc_task_instantiate(dentry, task, NULL);
3665out_drop_task:
3666 put_task_struct(task);
3667out:
3668 put_task_struct(leader);
3669out_no_task:
3670 return result;
3671}
3672
3673/*
3674 * Find the first tid of a thread group to return to user space.
3675 *
3676 * Usually this is just the thread group leader, but if the users
3677 * buffer was too small or there was a seek into the middle of the
3678 * directory we have more work todo.
3679 *
3680 * In the case of a short read we start with find_task_by_pid.
3681 *
3682 * In the case of a seek we start with the leader and walk nr
3683 * threads past it.
3684 */
3685static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3686 struct pid_namespace *ns)
3687{
3688 struct task_struct *pos, *task;
3689 unsigned long nr = f_pos;
3690
3691 if (nr != f_pos) /* 32bit overflow? */
3692 return NULL;
3693
3694 rcu_read_lock();
3695 task = pid_task(pid, PIDTYPE_PID);
3696 if (!task)
3697 goto fail;
3698
3699 /* Attempt to start with the tid of a thread */
3700 if (tid && nr) {
3701 pos = find_task_by_pid_ns(tid, ns);
3702 if (pos && same_thread_group(pos, task))
3703 goto found;
3704 }
3705
3706 /* If nr exceeds the number of threads there is nothing todo */
3707 if (nr >= get_nr_threads(task))
3708 goto fail;
3709
3710 /* If we haven't found our starting place yet start
3711 * with the leader and walk nr threads forward.
3712 */
3713 pos = task = task->group_leader;
3714 do {
3715 if (!nr--)
3716 goto found;
3717 } while_each_thread(task, pos);
3718fail:
3719 pos = NULL;
3720 goto out;
3721found:
3722 get_task_struct(pos);
3723out:
3724 rcu_read_unlock();
3725 return pos;
3726}
3727
3728/*
3729 * Find the next thread in the thread list.
3730 * Return NULL if there is an error or no next thread.
3731 *
3732 * The reference to the input task_struct is released.
3733 */
3734static struct task_struct *next_tid(struct task_struct *start)
3735{
3736 struct task_struct *pos = NULL;
3737 rcu_read_lock();
3738 if (pid_alive(start)) {
3739 pos = next_thread(start);
3740 if (thread_group_leader(pos))
3741 pos = NULL;
3742 else
3743 get_task_struct(pos);
3744 }
3745 rcu_read_unlock();
3746 put_task_struct(start);
3747 return pos;
3748}
3749
3750/* for the /proc/TGID/task/ directories */
3751static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3752{
3753 struct inode *inode = file_inode(file);
3754 struct task_struct *task;
3755 struct pid_namespace *ns;
3756 int tid;
3757
3758 if (proc_inode_is_dead(inode))
3759 return -ENOENT;
3760
3761 if (!dir_emit_dots(file, ctx))
3762 return 0;
3763
3764 /* f_version caches the tgid value that the last readdir call couldn't
3765 * return. lseek aka telldir automagically resets f_version to 0.
3766 */
3767 ns = proc_pid_ns(inode->i_sb);
3768 tid = (int)file->f_version;
3769 file->f_version = 0;
3770 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3771 task;
3772 task = next_tid(task), ctx->pos++) {
3773 char name[10 + 1];
3774 unsigned int len;
3775 tid = task_pid_nr_ns(task, ns);
3776 len = snprintf(name, sizeof(name), "%u", tid);
3777 if (!proc_fill_cache(file, ctx, name, len,
3778 proc_task_instantiate, task, NULL)) {
3779 /* returning this tgid failed, save it as the first
3780 * pid for the next readir call */
3781 file->f_version = (u64)tid;
3782 put_task_struct(task);
3783 break;
3784 }
3785 }
3786
3787 return 0;
3788}
3789
3790static int proc_task_getattr(const struct path *path, struct kstat *stat,
3791 u32 request_mask, unsigned int query_flags)
3792{
3793 struct inode *inode = d_inode(path->dentry);
3794 struct task_struct *p = get_proc_task(inode);
3795 generic_fillattr(inode, stat);
3796
3797 if (p) {
3798 stat->nlink += get_nr_threads(p);
3799 put_task_struct(p);
3800 }
3801
3802 return 0;
3803}
3804
3805static const struct inode_operations proc_task_inode_operations = {
3806 .lookup = proc_task_lookup,
3807 .getattr = proc_task_getattr,
3808 .setattr = proc_setattr,
3809 .permission = proc_pid_permission,
3810};
3811
3812static const struct file_operations proc_task_operations = {
3813 .read = generic_read_dir,
3814 .iterate_shared = proc_task_readdir,
3815 .llseek = generic_file_llseek,
3816};
3817
3818void __init set_proc_pid_nlink(void)
3819{
3820 nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3821 nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3822}
1/*
2 * linux/fs/proc/base.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * proc base directory handling functions
7 *
8 * 1999, Al Viro. Rewritten. Now it covers the whole per-process part.
9 * Instead of using magical inumbers to determine the kind of object
10 * we allocate and fill in-core inodes upon lookup. They don't even
11 * go into icache. We cache the reference to task_struct upon lookup too.
12 * Eventually it should become a filesystem in its own. We don't use the
13 * rest of procfs anymore.
14 *
15 *
16 * Changelog:
17 * 17-Jan-2005
18 * Allan Bezerra
19 * Bruna Moreira <bruna.moreira@indt.org.br>
20 * Edjard Mota <edjard.mota@indt.org.br>
21 * Ilias Biris <ilias.biris@indt.org.br>
22 * Mauricio Lin <mauricio.lin@indt.org.br>
23 *
24 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
25 *
26 * A new process specific entry (smaps) included in /proc. It shows the
27 * size of rss for each memory area. The maps entry lacks information
28 * about physical memory size (rss) for each mapped file, i.e.,
29 * rss information for executables and library files.
30 * This additional information is useful for any tools that need to know
31 * about physical memory consumption for a process specific library.
32 *
33 * Changelog:
34 * 21-Feb-2005
35 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
36 * Pud inclusion in the page table walking.
37 *
38 * ChangeLog:
39 * 10-Mar-2005
40 * 10LE Instituto Nokia de Tecnologia - INdT:
41 * A better way to walks through the page table as suggested by Hugh Dickins.
42 *
43 * Simo Piiroinen <simo.piiroinen@nokia.com>:
44 * Smaps information related to shared, private, clean and dirty pages.
45 *
46 * Paul Mundt <paul.mundt@nokia.com>:
47 * Overall revision about smaps.
48 */
49
50#include <asm/uaccess.h>
51
52#include <linux/errno.h>
53#include <linux/time.h>
54#include <linux/proc_fs.h>
55#include <linux/stat.h>
56#include <linux/task_io_accounting_ops.h>
57#include <linux/init.h>
58#include <linux/capability.h>
59#include <linux/file.h>
60#include <linux/fdtable.h>
61#include <linux/string.h>
62#include <linux/seq_file.h>
63#include <linux/namei.h>
64#include <linux/mnt_namespace.h>
65#include <linux/mm.h>
66#include <linux/swap.h>
67#include <linux/rcupdate.h>
68#include <linux/kallsyms.h>
69#include <linux/stacktrace.h>
70#include <linux/resource.h>
71#include <linux/module.h>
72#include <linux/mount.h>
73#include <linux/security.h>
74#include <linux/ptrace.h>
75#include <linux/tracehook.h>
76#include <linux/cgroup.h>
77#include <linux/cpuset.h>
78#include <linux/audit.h>
79#include <linux/poll.h>
80#include <linux/nsproxy.h>
81#include <linux/oom.h>
82#include <linux/elf.h>
83#include <linux/pid_namespace.h>
84#include <linux/user_namespace.h>
85#include <linux/fs_struct.h>
86#include <linux/slab.h>
87#include <linux/flex_array.h>
88#ifdef CONFIG_HARDWALL
89#include <asm/hardwall.h>
90#endif
91#include <trace/events/oom.h>
92#include "internal.h"
93
94/* NOTE:
95 * Implementing inode permission operations in /proc is almost
96 * certainly an error. Permission checks need to happen during
97 * each system call not at open time. The reason is that most of
98 * what we wish to check for permissions in /proc varies at runtime.
99 *
100 * The classic example of a problem is opening file descriptors
101 * in /proc for a task before it execs a suid executable.
102 */
103
104struct pid_entry {
105 char *name;
106 int len;
107 umode_t mode;
108 const struct inode_operations *iop;
109 const struct file_operations *fop;
110 union proc_op op;
111};
112
113#define NOD(NAME, MODE, IOP, FOP, OP) { \
114 .name = (NAME), \
115 .len = sizeof(NAME) - 1, \
116 .mode = MODE, \
117 .iop = IOP, \
118 .fop = FOP, \
119 .op = OP, \
120}
121
122#define DIR(NAME, MODE, iops, fops) \
123 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
124#define LNK(NAME, get_link) \
125 NOD(NAME, (S_IFLNK|S_IRWXUGO), \
126 &proc_pid_link_inode_operations, NULL, \
127 { .proc_get_link = get_link } )
128#define REG(NAME, MODE, fops) \
129 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
130#define INF(NAME, MODE, read) \
131 NOD(NAME, (S_IFREG|(MODE)), \
132 NULL, &proc_info_file_operations, \
133 { .proc_read = read } )
134#define ONE(NAME, MODE, show) \
135 NOD(NAME, (S_IFREG|(MODE)), \
136 NULL, &proc_single_file_operations, \
137 { .proc_show = show } )
138
139static int proc_fd_permission(struct inode *inode, int mask);
140
141/*
142 * Count the number of hardlinks for the pid_entry table, excluding the .
143 * and .. links.
144 */
145static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
146 unsigned int n)
147{
148 unsigned int i;
149 unsigned int count;
150
151 count = 0;
152 for (i = 0; i < n; ++i) {
153 if (S_ISDIR(entries[i].mode))
154 ++count;
155 }
156
157 return count;
158}
159
160static int get_task_root(struct task_struct *task, struct path *root)
161{
162 int result = -ENOENT;
163
164 task_lock(task);
165 if (task->fs) {
166 get_fs_root(task->fs, root);
167 result = 0;
168 }
169 task_unlock(task);
170 return result;
171}
172
173static int proc_cwd_link(struct dentry *dentry, struct path *path)
174{
175 struct task_struct *task = get_proc_task(dentry->d_inode);
176 int result = -ENOENT;
177
178 if (task) {
179 task_lock(task);
180 if (task->fs) {
181 get_fs_pwd(task->fs, path);
182 result = 0;
183 }
184 task_unlock(task);
185 put_task_struct(task);
186 }
187 return result;
188}
189
190static int proc_root_link(struct dentry *dentry, struct path *path)
191{
192 struct task_struct *task = get_proc_task(dentry->d_inode);
193 int result = -ENOENT;
194
195 if (task) {
196 result = get_task_root(task, path);
197 put_task_struct(task);
198 }
199 return result;
200}
201
202static int proc_pid_cmdline(struct task_struct *task, char * buffer)
203{
204 int res = 0;
205 unsigned int len;
206 struct mm_struct *mm = get_task_mm(task);
207 if (!mm)
208 goto out;
209 if (!mm->arg_end)
210 goto out_mm; /* Shh! No looking before we're done */
211
212 len = mm->arg_end - mm->arg_start;
213
214 if (len > PAGE_SIZE)
215 len = PAGE_SIZE;
216
217 res = access_process_vm(task, mm->arg_start, buffer, len, 0);
218
219 // If the nul at the end of args has been overwritten, then
220 // assume application is using setproctitle(3).
221 if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
222 len = strnlen(buffer, res);
223 if (len < res) {
224 res = len;
225 } else {
226 len = mm->env_end - mm->env_start;
227 if (len > PAGE_SIZE - res)
228 len = PAGE_SIZE - res;
229 res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
230 res = strnlen(buffer, res);
231 }
232 }
233out_mm:
234 mmput(mm);
235out:
236 return res;
237}
238
239static int proc_pid_auxv(struct task_struct *task, char *buffer)
240{
241 struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
242 int res = PTR_ERR(mm);
243 if (mm && !IS_ERR(mm)) {
244 unsigned int nwords = 0;
245 do {
246 nwords += 2;
247 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
248 res = nwords * sizeof(mm->saved_auxv[0]);
249 if (res > PAGE_SIZE)
250 res = PAGE_SIZE;
251 memcpy(buffer, mm->saved_auxv, res);
252 mmput(mm);
253 }
254 return res;
255}
256
257
258#ifdef CONFIG_KALLSYMS
259/*
260 * Provides a wchan file via kallsyms in a proper one-value-per-file format.
261 * Returns the resolved symbol. If that fails, simply return the address.
262 */
263static int proc_pid_wchan(struct task_struct *task, char *buffer)
264{
265 unsigned long wchan;
266 char symname[KSYM_NAME_LEN];
267
268 wchan = get_wchan(task);
269
270 if (lookup_symbol_name(wchan, symname) < 0)
271 if (!ptrace_may_access(task, PTRACE_MODE_READ))
272 return 0;
273 else
274 return sprintf(buffer, "%lu", wchan);
275 else
276 return sprintf(buffer, "%s", symname);
277}
278#endif /* CONFIG_KALLSYMS */
279
280static int lock_trace(struct task_struct *task)
281{
282 int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
283 if (err)
284 return err;
285 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
286 mutex_unlock(&task->signal->cred_guard_mutex);
287 return -EPERM;
288 }
289 return 0;
290}
291
292static void unlock_trace(struct task_struct *task)
293{
294 mutex_unlock(&task->signal->cred_guard_mutex);
295}
296
297#ifdef CONFIG_STACKTRACE
298
299#define MAX_STACK_TRACE_DEPTH 64
300
301static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
302 struct pid *pid, struct task_struct *task)
303{
304 struct stack_trace trace;
305 unsigned long *entries;
306 int err;
307 int i;
308
309 entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
310 if (!entries)
311 return -ENOMEM;
312
313 trace.nr_entries = 0;
314 trace.max_entries = MAX_STACK_TRACE_DEPTH;
315 trace.entries = entries;
316 trace.skip = 0;
317
318 err = lock_trace(task);
319 if (!err) {
320 save_stack_trace_tsk(task, &trace);
321
322 for (i = 0; i < trace.nr_entries; i++) {
323 seq_printf(m, "[<%pK>] %pS\n",
324 (void *)entries[i], (void *)entries[i]);
325 }
326 unlock_trace(task);
327 }
328 kfree(entries);
329
330 return err;
331}
332#endif
333
334#ifdef CONFIG_SCHEDSTATS
335/*
336 * Provides /proc/PID/schedstat
337 */
338static int proc_pid_schedstat(struct task_struct *task, char *buffer)
339{
340 return sprintf(buffer, "%llu %llu %lu\n",
341 (unsigned long long)task->se.sum_exec_runtime,
342 (unsigned long long)task->sched_info.run_delay,
343 task->sched_info.pcount);
344}
345#endif
346
347#ifdef CONFIG_LATENCYTOP
348static int lstats_show_proc(struct seq_file *m, void *v)
349{
350 int i;
351 struct inode *inode = m->private;
352 struct task_struct *task = get_proc_task(inode);
353
354 if (!task)
355 return -ESRCH;
356 seq_puts(m, "Latency Top version : v0.1\n");
357 for (i = 0; i < 32; i++) {
358 struct latency_record *lr = &task->latency_record[i];
359 if (lr->backtrace[0]) {
360 int q;
361 seq_printf(m, "%i %li %li",
362 lr->count, lr->time, lr->max);
363 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
364 unsigned long bt = lr->backtrace[q];
365 if (!bt)
366 break;
367 if (bt == ULONG_MAX)
368 break;
369 seq_printf(m, " %ps", (void *)bt);
370 }
371 seq_putc(m, '\n');
372 }
373
374 }
375 put_task_struct(task);
376 return 0;
377}
378
379static int lstats_open(struct inode *inode, struct file *file)
380{
381 return single_open(file, lstats_show_proc, inode);
382}
383
384static ssize_t lstats_write(struct file *file, const char __user *buf,
385 size_t count, loff_t *offs)
386{
387 struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
388
389 if (!task)
390 return -ESRCH;
391 clear_all_latency_tracing(task);
392 put_task_struct(task);
393
394 return count;
395}
396
397static const struct file_operations proc_lstats_operations = {
398 .open = lstats_open,
399 .read = seq_read,
400 .write = lstats_write,
401 .llseek = seq_lseek,
402 .release = single_release,
403};
404
405#endif
406
407static int proc_oom_score(struct task_struct *task, char *buffer)
408{
409 unsigned long totalpages = totalram_pages + total_swap_pages;
410 unsigned long points = 0;
411
412 read_lock(&tasklist_lock);
413 if (pid_alive(task))
414 points = oom_badness(task, NULL, NULL, totalpages) *
415 1000 / totalpages;
416 read_unlock(&tasklist_lock);
417 return sprintf(buffer, "%lu\n", points);
418}
419
420struct limit_names {
421 char *name;
422 char *unit;
423};
424
425static const struct limit_names lnames[RLIM_NLIMITS] = {
426 [RLIMIT_CPU] = {"Max cpu time", "seconds"},
427 [RLIMIT_FSIZE] = {"Max file size", "bytes"},
428 [RLIMIT_DATA] = {"Max data size", "bytes"},
429 [RLIMIT_STACK] = {"Max stack size", "bytes"},
430 [RLIMIT_CORE] = {"Max core file size", "bytes"},
431 [RLIMIT_RSS] = {"Max resident set", "bytes"},
432 [RLIMIT_NPROC] = {"Max processes", "processes"},
433 [RLIMIT_NOFILE] = {"Max open files", "files"},
434 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
435 [RLIMIT_AS] = {"Max address space", "bytes"},
436 [RLIMIT_LOCKS] = {"Max file locks", "locks"},
437 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
438 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
439 [RLIMIT_NICE] = {"Max nice priority", NULL},
440 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
441 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
442};
443
444/* Display limits for a process */
445static int proc_pid_limits(struct task_struct *task, char *buffer)
446{
447 unsigned int i;
448 int count = 0;
449 unsigned long flags;
450 char *bufptr = buffer;
451
452 struct rlimit rlim[RLIM_NLIMITS];
453
454 if (!lock_task_sighand(task, &flags))
455 return 0;
456 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
457 unlock_task_sighand(task, &flags);
458
459 /*
460 * print the file header
461 */
462 count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
463 "Limit", "Soft Limit", "Hard Limit", "Units");
464
465 for (i = 0; i < RLIM_NLIMITS; i++) {
466 if (rlim[i].rlim_cur == RLIM_INFINITY)
467 count += sprintf(&bufptr[count], "%-25s %-20s ",
468 lnames[i].name, "unlimited");
469 else
470 count += sprintf(&bufptr[count], "%-25s %-20lu ",
471 lnames[i].name, rlim[i].rlim_cur);
472
473 if (rlim[i].rlim_max == RLIM_INFINITY)
474 count += sprintf(&bufptr[count], "%-20s ", "unlimited");
475 else
476 count += sprintf(&bufptr[count], "%-20lu ",
477 rlim[i].rlim_max);
478
479 if (lnames[i].unit)
480 count += sprintf(&bufptr[count], "%-10s\n",
481 lnames[i].unit);
482 else
483 count += sprintf(&bufptr[count], "\n");
484 }
485
486 return count;
487}
488
489#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
490static int proc_pid_syscall(struct task_struct *task, char *buffer)
491{
492 long nr;
493 unsigned long args[6], sp, pc;
494 int res = lock_trace(task);
495 if (res)
496 return res;
497
498 if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
499 res = sprintf(buffer, "running\n");
500 else if (nr < 0)
501 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
502 else
503 res = sprintf(buffer,
504 "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
505 nr,
506 args[0], args[1], args[2], args[3], args[4], args[5],
507 sp, pc);
508 unlock_trace(task);
509 return res;
510}
511#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
512
513/************************************************************************/
514/* Here the fs part begins */
515/************************************************************************/
516
517/* permission checks */
518static int proc_fd_access_allowed(struct inode *inode)
519{
520 struct task_struct *task;
521 int allowed = 0;
522 /* Allow access to a task's file descriptors if it is us or we
523 * may use ptrace attach to the process and find out that
524 * information.
525 */
526 task = get_proc_task(inode);
527 if (task) {
528 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
529 put_task_struct(task);
530 }
531 return allowed;
532}
533
534int proc_setattr(struct dentry *dentry, struct iattr *attr)
535{
536 int error;
537 struct inode *inode = dentry->d_inode;
538
539 if (attr->ia_valid & ATTR_MODE)
540 return -EPERM;
541
542 error = inode_change_ok(inode, attr);
543 if (error)
544 return error;
545
546 if ((attr->ia_valid & ATTR_SIZE) &&
547 attr->ia_size != i_size_read(inode)) {
548 error = vmtruncate(inode, attr->ia_size);
549 if (error)
550 return error;
551 }
552
553 setattr_copy(inode, attr);
554 mark_inode_dirty(inode);
555 return 0;
556}
557
558/*
559 * May current process learn task's sched/cmdline info (for hide_pid_min=1)
560 * or euid/egid (for hide_pid_min=2)?
561 */
562static bool has_pid_permissions(struct pid_namespace *pid,
563 struct task_struct *task,
564 int hide_pid_min)
565{
566 if (pid->hide_pid < hide_pid_min)
567 return true;
568 if (in_group_p(pid->pid_gid))
569 return true;
570 return ptrace_may_access(task, PTRACE_MODE_READ);
571}
572
573
574static int proc_pid_permission(struct inode *inode, int mask)
575{
576 struct pid_namespace *pid = inode->i_sb->s_fs_info;
577 struct task_struct *task;
578 bool has_perms;
579
580 task = get_proc_task(inode);
581 if (!task)
582 return -ESRCH;
583 has_perms = has_pid_permissions(pid, task, 1);
584 put_task_struct(task);
585
586 if (!has_perms) {
587 if (pid->hide_pid == 2) {
588 /*
589 * Let's make getdents(), stat(), and open()
590 * consistent with each other. If a process
591 * may not stat() a file, it shouldn't be seen
592 * in procfs at all.
593 */
594 return -ENOENT;
595 }
596
597 return -EPERM;
598 }
599 return generic_permission(inode, mask);
600}
601
602
603
604static const struct inode_operations proc_def_inode_operations = {
605 .setattr = proc_setattr,
606};
607
608#define PROC_BLOCK_SIZE (3*1024) /* 4K page size but our output routines use some slack for overruns */
609
610static ssize_t proc_info_read(struct file * file, char __user * buf,
611 size_t count, loff_t *ppos)
612{
613 struct inode * inode = file->f_path.dentry->d_inode;
614 unsigned long page;
615 ssize_t length;
616 struct task_struct *task = get_proc_task(inode);
617
618 length = -ESRCH;
619 if (!task)
620 goto out_no_task;
621
622 if (count > PROC_BLOCK_SIZE)
623 count = PROC_BLOCK_SIZE;
624
625 length = -ENOMEM;
626 if (!(page = __get_free_page(GFP_TEMPORARY)))
627 goto out;
628
629 length = PROC_I(inode)->op.proc_read(task, (char*)page);
630
631 if (length >= 0)
632 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
633 free_page(page);
634out:
635 put_task_struct(task);
636out_no_task:
637 return length;
638}
639
640static const struct file_operations proc_info_file_operations = {
641 .read = proc_info_read,
642 .llseek = generic_file_llseek,
643};
644
645static int proc_single_show(struct seq_file *m, void *v)
646{
647 struct inode *inode = m->private;
648 struct pid_namespace *ns;
649 struct pid *pid;
650 struct task_struct *task;
651 int ret;
652
653 ns = inode->i_sb->s_fs_info;
654 pid = proc_pid(inode);
655 task = get_pid_task(pid, PIDTYPE_PID);
656 if (!task)
657 return -ESRCH;
658
659 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
660
661 put_task_struct(task);
662 return ret;
663}
664
665static int proc_single_open(struct inode *inode, struct file *filp)
666{
667 return single_open(filp, proc_single_show, inode);
668}
669
670static const struct file_operations proc_single_file_operations = {
671 .open = proc_single_open,
672 .read = seq_read,
673 .llseek = seq_lseek,
674 .release = single_release,
675};
676
677static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
678{
679 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
680 struct mm_struct *mm;
681
682 if (!task)
683 return -ESRCH;
684
685 mm = mm_access(task, mode);
686 put_task_struct(task);
687
688 if (IS_ERR(mm))
689 return PTR_ERR(mm);
690
691 if (mm) {
692 /* ensure this mm_struct can't be freed */
693 atomic_inc(&mm->mm_count);
694 /* but do not pin its memory */
695 mmput(mm);
696 }
697
698 /* OK to pass negative loff_t, we can catch out-of-range */
699 file->f_mode |= FMODE_UNSIGNED_OFFSET;
700 file->private_data = mm;
701
702 return 0;
703}
704
705static int mem_open(struct inode *inode, struct file *file)
706{
707 return __mem_open(inode, file, PTRACE_MODE_ATTACH);
708}
709
710static ssize_t mem_rw(struct file *file, char __user *buf,
711 size_t count, loff_t *ppos, int write)
712{
713 struct mm_struct *mm = file->private_data;
714 unsigned long addr = *ppos;
715 ssize_t copied;
716 char *page;
717
718 if (!mm)
719 return 0;
720
721 page = (char *)__get_free_page(GFP_TEMPORARY);
722 if (!page)
723 return -ENOMEM;
724
725 copied = 0;
726 if (!atomic_inc_not_zero(&mm->mm_users))
727 goto free;
728
729 while (count > 0) {
730 int this_len = min_t(int, count, PAGE_SIZE);
731
732 if (write && copy_from_user(page, buf, this_len)) {
733 copied = -EFAULT;
734 break;
735 }
736
737 this_len = access_remote_vm(mm, addr, page, this_len, write);
738 if (!this_len) {
739 if (!copied)
740 copied = -EIO;
741 break;
742 }
743
744 if (!write && copy_to_user(buf, page, this_len)) {
745 copied = -EFAULT;
746 break;
747 }
748
749 buf += this_len;
750 addr += this_len;
751 copied += this_len;
752 count -= this_len;
753 }
754 *ppos = addr;
755
756 mmput(mm);
757free:
758 free_page((unsigned long) page);
759 return copied;
760}
761
762static ssize_t mem_read(struct file *file, char __user *buf,
763 size_t count, loff_t *ppos)
764{
765 return mem_rw(file, buf, count, ppos, 0);
766}
767
768static ssize_t mem_write(struct file *file, const char __user *buf,
769 size_t count, loff_t *ppos)
770{
771 return mem_rw(file, (char __user*)buf, count, ppos, 1);
772}
773
774loff_t mem_lseek(struct file *file, loff_t offset, int orig)
775{
776 switch (orig) {
777 case 0:
778 file->f_pos = offset;
779 break;
780 case 1:
781 file->f_pos += offset;
782 break;
783 default:
784 return -EINVAL;
785 }
786 force_successful_syscall_return();
787 return file->f_pos;
788}
789
790static int mem_release(struct inode *inode, struct file *file)
791{
792 struct mm_struct *mm = file->private_data;
793 if (mm)
794 mmdrop(mm);
795 return 0;
796}
797
798static const struct file_operations proc_mem_operations = {
799 .llseek = mem_lseek,
800 .read = mem_read,
801 .write = mem_write,
802 .open = mem_open,
803 .release = mem_release,
804};
805
806static int environ_open(struct inode *inode, struct file *file)
807{
808 return __mem_open(inode, file, PTRACE_MODE_READ);
809}
810
811static ssize_t environ_read(struct file *file, char __user *buf,
812 size_t count, loff_t *ppos)
813{
814 char *page;
815 unsigned long src = *ppos;
816 int ret = 0;
817 struct mm_struct *mm = file->private_data;
818
819 if (!mm)
820 return 0;
821
822 page = (char *)__get_free_page(GFP_TEMPORARY);
823 if (!page)
824 return -ENOMEM;
825
826 ret = 0;
827 if (!atomic_inc_not_zero(&mm->mm_users))
828 goto free;
829 while (count > 0) {
830 int this_len, retval, max_len;
831
832 this_len = mm->env_end - (mm->env_start + src);
833
834 if (this_len <= 0)
835 break;
836
837 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
838 this_len = (this_len > max_len) ? max_len : this_len;
839
840 retval = access_remote_vm(mm, (mm->env_start + src),
841 page, this_len, 0);
842
843 if (retval <= 0) {
844 ret = retval;
845 break;
846 }
847
848 if (copy_to_user(buf, page, retval)) {
849 ret = -EFAULT;
850 break;
851 }
852
853 ret += retval;
854 src += retval;
855 buf += retval;
856 count -= retval;
857 }
858 *ppos = src;
859 mmput(mm);
860
861free:
862 free_page((unsigned long) page);
863 return ret;
864}
865
866static const struct file_operations proc_environ_operations = {
867 .open = environ_open,
868 .read = environ_read,
869 .llseek = generic_file_llseek,
870 .release = mem_release,
871};
872
873static ssize_t oom_adjust_read(struct file *file, char __user *buf,
874 size_t count, loff_t *ppos)
875{
876 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
877 char buffer[PROC_NUMBUF];
878 size_t len;
879 int oom_adjust = OOM_DISABLE;
880 unsigned long flags;
881
882 if (!task)
883 return -ESRCH;
884
885 if (lock_task_sighand(task, &flags)) {
886 oom_adjust = task->signal->oom_adj;
887 unlock_task_sighand(task, &flags);
888 }
889
890 put_task_struct(task);
891
892 len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
893
894 return simple_read_from_buffer(buf, count, ppos, buffer, len);
895}
896
897static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
898 size_t count, loff_t *ppos)
899{
900 struct task_struct *task;
901 char buffer[PROC_NUMBUF];
902 int oom_adjust;
903 unsigned long flags;
904 int err;
905
906 memset(buffer, 0, sizeof(buffer));
907 if (count > sizeof(buffer) - 1)
908 count = sizeof(buffer) - 1;
909 if (copy_from_user(buffer, buf, count)) {
910 err = -EFAULT;
911 goto out;
912 }
913
914 err = kstrtoint(strstrip(buffer), 0, &oom_adjust);
915 if (err)
916 goto out;
917 if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
918 oom_adjust != OOM_DISABLE) {
919 err = -EINVAL;
920 goto out;
921 }
922
923 task = get_proc_task(file->f_path.dentry->d_inode);
924 if (!task) {
925 err = -ESRCH;
926 goto out;
927 }
928
929 task_lock(task);
930 if (!task->mm) {
931 err = -EINVAL;
932 goto err_task_lock;
933 }
934
935 if (!lock_task_sighand(task, &flags)) {
936 err = -ESRCH;
937 goto err_task_lock;
938 }
939
940 if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
941 err = -EACCES;
942 goto err_sighand;
943 }
944
945 /*
946 * Warn that /proc/pid/oom_adj is deprecated, see
947 * Documentation/feature-removal-schedule.txt.
948 */
949 printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
950 current->comm, task_pid_nr(current), task_pid_nr(task),
951 task_pid_nr(task));
952 task->signal->oom_adj = oom_adjust;
953 /*
954 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
955 * value is always attainable.
956 */
957 if (task->signal->oom_adj == OOM_ADJUST_MAX)
958 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
959 else
960 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
961 -OOM_DISABLE;
962 trace_oom_score_adj_update(task);
963err_sighand:
964 unlock_task_sighand(task, &flags);
965err_task_lock:
966 task_unlock(task);
967 put_task_struct(task);
968out:
969 return err < 0 ? err : count;
970}
971
972static const struct file_operations proc_oom_adjust_operations = {
973 .read = oom_adjust_read,
974 .write = oom_adjust_write,
975 .llseek = generic_file_llseek,
976};
977
978static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
979 size_t count, loff_t *ppos)
980{
981 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
982 char buffer[PROC_NUMBUF];
983 int oom_score_adj = OOM_SCORE_ADJ_MIN;
984 unsigned long flags;
985 size_t len;
986
987 if (!task)
988 return -ESRCH;
989 if (lock_task_sighand(task, &flags)) {
990 oom_score_adj = task->signal->oom_score_adj;
991 unlock_task_sighand(task, &flags);
992 }
993 put_task_struct(task);
994 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
995 return simple_read_from_buffer(buf, count, ppos, buffer, len);
996}
997
998static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
999 size_t count, loff_t *ppos)
1000{
1001 struct task_struct *task;
1002 char buffer[PROC_NUMBUF];
1003 unsigned long flags;
1004 int oom_score_adj;
1005 int err;
1006
1007 memset(buffer, 0, sizeof(buffer));
1008 if (count > sizeof(buffer) - 1)
1009 count = sizeof(buffer) - 1;
1010 if (copy_from_user(buffer, buf, count)) {
1011 err = -EFAULT;
1012 goto out;
1013 }
1014
1015 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1016 if (err)
1017 goto out;
1018 if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1019 oom_score_adj > OOM_SCORE_ADJ_MAX) {
1020 err = -EINVAL;
1021 goto out;
1022 }
1023
1024 task = get_proc_task(file->f_path.dentry->d_inode);
1025 if (!task) {
1026 err = -ESRCH;
1027 goto out;
1028 }
1029
1030 task_lock(task);
1031 if (!task->mm) {
1032 err = -EINVAL;
1033 goto err_task_lock;
1034 }
1035
1036 if (!lock_task_sighand(task, &flags)) {
1037 err = -ESRCH;
1038 goto err_task_lock;
1039 }
1040
1041 if (oom_score_adj < task->signal->oom_score_adj_min &&
1042 !capable(CAP_SYS_RESOURCE)) {
1043 err = -EACCES;
1044 goto err_sighand;
1045 }
1046
1047 task->signal->oom_score_adj = oom_score_adj;
1048 if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1049 task->signal->oom_score_adj_min = oom_score_adj;
1050 trace_oom_score_adj_update(task);
1051 /*
1052 * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1053 * always attainable.
1054 */
1055 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1056 task->signal->oom_adj = OOM_DISABLE;
1057 else
1058 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1059 OOM_SCORE_ADJ_MAX;
1060err_sighand:
1061 unlock_task_sighand(task, &flags);
1062err_task_lock:
1063 task_unlock(task);
1064 put_task_struct(task);
1065out:
1066 return err < 0 ? err : count;
1067}
1068
1069static const struct file_operations proc_oom_score_adj_operations = {
1070 .read = oom_score_adj_read,
1071 .write = oom_score_adj_write,
1072 .llseek = default_llseek,
1073};
1074
1075#ifdef CONFIG_AUDITSYSCALL
1076#define TMPBUFLEN 21
1077static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1078 size_t count, loff_t *ppos)
1079{
1080 struct inode * inode = file->f_path.dentry->d_inode;
1081 struct task_struct *task = get_proc_task(inode);
1082 ssize_t length;
1083 char tmpbuf[TMPBUFLEN];
1084
1085 if (!task)
1086 return -ESRCH;
1087 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1088 audit_get_loginuid(task));
1089 put_task_struct(task);
1090 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1091}
1092
1093static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1094 size_t count, loff_t *ppos)
1095{
1096 struct inode * inode = file->f_path.dentry->d_inode;
1097 char *page, *tmp;
1098 ssize_t length;
1099 uid_t loginuid;
1100
1101 rcu_read_lock();
1102 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1103 rcu_read_unlock();
1104 return -EPERM;
1105 }
1106 rcu_read_unlock();
1107
1108 if (count >= PAGE_SIZE)
1109 count = PAGE_SIZE - 1;
1110
1111 if (*ppos != 0) {
1112 /* No partial writes. */
1113 return -EINVAL;
1114 }
1115 page = (char*)__get_free_page(GFP_TEMPORARY);
1116 if (!page)
1117 return -ENOMEM;
1118 length = -EFAULT;
1119 if (copy_from_user(page, buf, count))
1120 goto out_free_page;
1121
1122 page[count] = '\0';
1123 loginuid = simple_strtoul(page, &tmp, 10);
1124 if (tmp == page) {
1125 length = -EINVAL;
1126 goto out_free_page;
1127
1128 }
1129 length = audit_set_loginuid(loginuid);
1130 if (likely(length == 0))
1131 length = count;
1132
1133out_free_page:
1134 free_page((unsigned long) page);
1135 return length;
1136}
1137
1138static const struct file_operations proc_loginuid_operations = {
1139 .read = proc_loginuid_read,
1140 .write = proc_loginuid_write,
1141 .llseek = generic_file_llseek,
1142};
1143
1144static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1145 size_t count, loff_t *ppos)
1146{
1147 struct inode * inode = file->f_path.dentry->d_inode;
1148 struct task_struct *task = get_proc_task(inode);
1149 ssize_t length;
1150 char tmpbuf[TMPBUFLEN];
1151
1152 if (!task)
1153 return -ESRCH;
1154 length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1155 audit_get_sessionid(task));
1156 put_task_struct(task);
1157 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1158}
1159
1160static const struct file_operations proc_sessionid_operations = {
1161 .read = proc_sessionid_read,
1162 .llseek = generic_file_llseek,
1163};
1164#endif
1165
1166#ifdef CONFIG_FAULT_INJECTION
1167static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1168 size_t count, loff_t *ppos)
1169{
1170 struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1171 char buffer[PROC_NUMBUF];
1172 size_t len;
1173 int make_it_fail;
1174
1175 if (!task)
1176 return -ESRCH;
1177 make_it_fail = task->make_it_fail;
1178 put_task_struct(task);
1179
1180 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1181
1182 return simple_read_from_buffer(buf, count, ppos, buffer, len);
1183}
1184
1185static ssize_t proc_fault_inject_write(struct file * file,
1186 const char __user * buf, size_t count, loff_t *ppos)
1187{
1188 struct task_struct *task;
1189 char buffer[PROC_NUMBUF], *end;
1190 int make_it_fail;
1191
1192 if (!capable(CAP_SYS_RESOURCE))
1193 return -EPERM;
1194 memset(buffer, 0, sizeof(buffer));
1195 if (count > sizeof(buffer) - 1)
1196 count = sizeof(buffer) - 1;
1197 if (copy_from_user(buffer, buf, count))
1198 return -EFAULT;
1199 make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1200 if (*end)
1201 return -EINVAL;
1202 task = get_proc_task(file->f_dentry->d_inode);
1203 if (!task)
1204 return -ESRCH;
1205 task->make_it_fail = make_it_fail;
1206 put_task_struct(task);
1207
1208 return count;
1209}
1210
1211static const struct file_operations proc_fault_inject_operations = {
1212 .read = proc_fault_inject_read,
1213 .write = proc_fault_inject_write,
1214 .llseek = generic_file_llseek,
1215};
1216#endif
1217
1218
1219#ifdef CONFIG_SCHED_DEBUG
1220/*
1221 * Print out various scheduling related per-task fields:
1222 */
1223static int sched_show(struct seq_file *m, void *v)
1224{
1225 struct inode *inode = m->private;
1226 struct task_struct *p;
1227
1228 p = get_proc_task(inode);
1229 if (!p)
1230 return -ESRCH;
1231 proc_sched_show_task(p, m);
1232
1233 put_task_struct(p);
1234
1235 return 0;
1236}
1237
1238static ssize_t
1239sched_write(struct file *file, const char __user *buf,
1240 size_t count, loff_t *offset)
1241{
1242 struct inode *inode = file->f_path.dentry->d_inode;
1243 struct task_struct *p;
1244
1245 p = get_proc_task(inode);
1246 if (!p)
1247 return -ESRCH;
1248 proc_sched_set_task(p);
1249
1250 put_task_struct(p);
1251
1252 return count;
1253}
1254
1255static int sched_open(struct inode *inode, struct file *filp)
1256{
1257 return single_open(filp, sched_show, inode);
1258}
1259
1260static const struct file_operations proc_pid_sched_operations = {
1261 .open = sched_open,
1262 .read = seq_read,
1263 .write = sched_write,
1264 .llseek = seq_lseek,
1265 .release = single_release,
1266};
1267
1268#endif
1269
1270#ifdef CONFIG_SCHED_AUTOGROUP
1271/*
1272 * Print out autogroup related information:
1273 */
1274static int sched_autogroup_show(struct seq_file *m, void *v)
1275{
1276 struct inode *inode = m->private;
1277 struct task_struct *p;
1278
1279 p = get_proc_task(inode);
1280 if (!p)
1281 return -ESRCH;
1282 proc_sched_autogroup_show_task(p, m);
1283
1284 put_task_struct(p);
1285
1286 return 0;
1287}
1288
1289static ssize_t
1290sched_autogroup_write(struct file *file, const char __user *buf,
1291 size_t count, loff_t *offset)
1292{
1293 struct inode *inode = file->f_path.dentry->d_inode;
1294 struct task_struct *p;
1295 char buffer[PROC_NUMBUF];
1296 int nice;
1297 int err;
1298
1299 memset(buffer, 0, sizeof(buffer));
1300 if (count > sizeof(buffer) - 1)
1301 count = sizeof(buffer) - 1;
1302 if (copy_from_user(buffer, buf, count))
1303 return -EFAULT;
1304
1305 err = kstrtoint(strstrip(buffer), 0, &nice);
1306 if (err < 0)
1307 return err;
1308
1309 p = get_proc_task(inode);
1310 if (!p)
1311 return -ESRCH;
1312
1313 err = proc_sched_autogroup_set_nice(p, nice);
1314 if (err)
1315 count = err;
1316
1317 put_task_struct(p);
1318
1319 return count;
1320}
1321
1322static int sched_autogroup_open(struct inode *inode, struct file *filp)
1323{
1324 int ret;
1325
1326 ret = single_open(filp, sched_autogroup_show, NULL);
1327 if (!ret) {
1328 struct seq_file *m = filp->private_data;
1329
1330 m->private = inode;
1331 }
1332 return ret;
1333}
1334
1335static const struct file_operations proc_pid_sched_autogroup_operations = {
1336 .open = sched_autogroup_open,
1337 .read = seq_read,
1338 .write = sched_autogroup_write,
1339 .llseek = seq_lseek,
1340 .release = single_release,
1341};
1342
1343#endif /* CONFIG_SCHED_AUTOGROUP */
1344
1345static ssize_t comm_write(struct file *file, const char __user *buf,
1346 size_t count, loff_t *offset)
1347{
1348 struct inode *inode = file->f_path.dentry->d_inode;
1349 struct task_struct *p;
1350 char buffer[TASK_COMM_LEN];
1351
1352 memset(buffer, 0, sizeof(buffer));
1353 if (count > sizeof(buffer) - 1)
1354 count = sizeof(buffer) - 1;
1355 if (copy_from_user(buffer, buf, count))
1356 return -EFAULT;
1357
1358 p = get_proc_task(inode);
1359 if (!p)
1360 return -ESRCH;
1361
1362 if (same_thread_group(current, p))
1363 set_task_comm(p, buffer);
1364 else
1365 count = -EINVAL;
1366
1367 put_task_struct(p);
1368
1369 return count;
1370}
1371
1372static int comm_show(struct seq_file *m, void *v)
1373{
1374 struct inode *inode = m->private;
1375 struct task_struct *p;
1376
1377 p = get_proc_task(inode);
1378 if (!p)
1379 return -ESRCH;
1380
1381 task_lock(p);
1382 seq_printf(m, "%s\n", p->comm);
1383 task_unlock(p);
1384
1385 put_task_struct(p);
1386
1387 return 0;
1388}
1389
1390static int comm_open(struct inode *inode, struct file *filp)
1391{
1392 return single_open(filp, comm_show, inode);
1393}
1394
1395static const struct file_operations proc_pid_set_comm_operations = {
1396 .open = comm_open,
1397 .read = seq_read,
1398 .write = comm_write,
1399 .llseek = seq_lseek,
1400 .release = single_release,
1401};
1402
1403static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1404{
1405 struct task_struct *task;
1406 struct mm_struct *mm;
1407 struct file *exe_file;
1408
1409 task = get_proc_task(dentry->d_inode);
1410 if (!task)
1411 return -ENOENT;
1412 mm = get_task_mm(task);
1413 put_task_struct(task);
1414 if (!mm)
1415 return -ENOENT;
1416 exe_file = get_mm_exe_file(mm);
1417 mmput(mm);
1418 if (exe_file) {
1419 *exe_path = exe_file->f_path;
1420 path_get(&exe_file->f_path);
1421 fput(exe_file);
1422 return 0;
1423 } else
1424 return -ENOENT;
1425}
1426
1427static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1428{
1429 struct inode *inode = dentry->d_inode;
1430 int error = -EACCES;
1431
1432 /* We don't need a base pointer in the /proc filesystem */
1433 path_put(&nd->path);
1434
1435 /* Are we allowed to snoop on the tasks file descriptors? */
1436 if (!proc_fd_access_allowed(inode))
1437 goto out;
1438
1439 error = PROC_I(inode)->op.proc_get_link(dentry, &nd->path);
1440out:
1441 return ERR_PTR(error);
1442}
1443
1444static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1445{
1446 char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1447 char *pathname;
1448 int len;
1449
1450 if (!tmp)
1451 return -ENOMEM;
1452
1453 pathname = d_path(path, tmp, PAGE_SIZE);
1454 len = PTR_ERR(pathname);
1455 if (IS_ERR(pathname))
1456 goto out;
1457 len = tmp + PAGE_SIZE - 1 - pathname;
1458
1459 if (len > buflen)
1460 len = buflen;
1461 if (copy_to_user(buffer, pathname, len))
1462 len = -EFAULT;
1463 out:
1464 free_page((unsigned long)tmp);
1465 return len;
1466}
1467
1468static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1469{
1470 int error = -EACCES;
1471 struct inode *inode = dentry->d_inode;
1472 struct path path;
1473
1474 /* Are we allowed to snoop on the tasks file descriptors? */
1475 if (!proc_fd_access_allowed(inode))
1476 goto out;
1477
1478 error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1479 if (error)
1480 goto out;
1481
1482 error = do_proc_readlink(&path, buffer, buflen);
1483 path_put(&path);
1484out:
1485 return error;
1486}
1487
1488static const struct inode_operations proc_pid_link_inode_operations = {
1489 .readlink = proc_pid_readlink,
1490 .follow_link = proc_pid_follow_link,
1491 .setattr = proc_setattr,
1492};
1493
1494
1495/* building an inode */
1496
1497static int task_dumpable(struct task_struct *task)
1498{
1499 int dumpable = 0;
1500 struct mm_struct *mm;
1501
1502 task_lock(task);
1503 mm = task->mm;
1504 if (mm)
1505 dumpable = get_dumpable(mm);
1506 task_unlock(task);
1507 if(dumpable == 1)
1508 return 1;
1509 return 0;
1510}
1511
1512struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1513{
1514 struct inode * inode;
1515 struct proc_inode *ei;
1516 const struct cred *cred;
1517
1518 /* We need a new inode */
1519
1520 inode = new_inode(sb);
1521 if (!inode)
1522 goto out;
1523
1524 /* Common stuff */
1525 ei = PROC_I(inode);
1526 inode->i_ino = get_next_ino();
1527 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1528 inode->i_op = &proc_def_inode_operations;
1529
1530 /*
1531 * grab the reference to task.
1532 */
1533 ei->pid = get_task_pid(task, PIDTYPE_PID);
1534 if (!ei->pid)
1535 goto out_unlock;
1536
1537 if (task_dumpable(task)) {
1538 rcu_read_lock();
1539 cred = __task_cred(task);
1540 inode->i_uid = cred->euid;
1541 inode->i_gid = cred->egid;
1542 rcu_read_unlock();
1543 }
1544 security_task_to_inode(task, inode);
1545
1546out:
1547 return inode;
1548
1549out_unlock:
1550 iput(inode);
1551 return NULL;
1552}
1553
1554int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1555{
1556 struct inode *inode = dentry->d_inode;
1557 struct task_struct *task;
1558 const struct cred *cred;
1559 struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1560
1561 generic_fillattr(inode, stat);
1562
1563 rcu_read_lock();
1564 stat->uid = GLOBAL_ROOT_UID;
1565 stat->gid = GLOBAL_ROOT_GID;
1566 task = pid_task(proc_pid(inode), PIDTYPE_PID);
1567 if (task) {
1568 if (!has_pid_permissions(pid, task, 2)) {
1569 rcu_read_unlock();
1570 /*
1571 * This doesn't prevent learning whether PID exists,
1572 * it only makes getattr() consistent with readdir().
1573 */
1574 return -ENOENT;
1575 }
1576 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1577 task_dumpable(task)) {
1578 cred = __task_cred(task);
1579 stat->uid = cred->euid;
1580 stat->gid = cred->egid;
1581 }
1582 }
1583 rcu_read_unlock();
1584 return 0;
1585}
1586
1587/* dentry stuff */
1588
1589/*
1590 * Exceptional case: normally we are not allowed to unhash a busy
1591 * directory. In this case, however, we can do it - no aliasing problems
1592 * due to the way we treat inodes.
1593 *
1594 * Rewrite the inode's ownerships here because the owning task may have
1595 * performed a setuid(), etc.
1596 *
1597 * Before the /proc/pid/status file was created the only way to read
1598 * the effective uid of a /process was to stat /proc/pid. Reading
1599 * /proc/pid/status is slow enough that procps and other packages
1600 * kept stating /proc/pid. To keep the rules in /proc simple I have
1601 * made this apply to all per process world readable and executable
1602 * directories.
1603 */
1604int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1605{
1606 struct inode *inode;
1607 struct task_struct *task;
1608 const struct cred *cred;
1609
1610 if (nd && nd->flags & LOOKUP_RCU)
1611 return -ECHILD;
1612
1613 inode = dentry->d_inode;
1614 task = get_proc_task(inode);
1615
1616 if (task) {
1617 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1618 task_dumpable(task)) {
1619 rcu_read_lock();
1620 cred = __task_cred(task);
1621 inode->i_uid = cred->euid;
1622 inode->i_gid = cred->egid;
1623 rcu_read_unlock();
1624 } else {
1625 inode->i_uid = GLOBAL_ROOT_UID;
1626 inode->i_gid = GLOBAL_ROOT_GID;
1627 }
1628 inode->i_mode &= ~(S_ISUID | S_ISGID);
1629 security_task_to_inode(task, inode);
1630 put_task_struct(task);
1631 return 1;
1632 }
1633 d_drop(dentry);
1634 return 0;
1635}
1636
1637static int pid_delete_dentry(const struct dentry * dentry)
1638{
1639 /* Is the task we represent dead?
1640 * If so, then don't put the dentry on the lru list,
1641 * kill it immediately.
1642 */
1643 return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1644}
1645
1646const struct dentry_operations pid_dentry_operations =
1647{
1648 .d_revalidate = pid_revalidate,
1649 .d_delete = pid_delete_dentry,
1650};
1651
1652/* Lookups */
1653
1654/*
1655 * Fill a directory entry.
1656 *
1657 * If possible create the dcache entry and derive our inode number and
1658 * file type from dcache entry.
1659 *
1660 * Since all of the proc inode numbers are dynamically generated, the inode
1661 * numbers do not exist until the inode is cache. This means creating the
1662 * the dcache entry in readdir is necessary to keep the inode numbers
1663 * reported by readdir in sync with the inode numbers reported
1664 * by stat.
1665 */
1666int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1667 const char *name, int len,
1668 instantiate_t instantiate, struct task_struct *task, const void *ptr)
1669{
1670 struct dentry *child, *dir = filp->f_path.dentry;
1671 struct inode *inode;
1672 struct qstr qname;
1673 ino_t ino = 0;
1674 unsigned type = DT_UNKNOWN;
1675
1676 qname.name = name;
1677 qname.len = len;
1678 qname.hash = full_name_hash(name, len);
1679
1680 child = d_lookup(dir, &qname);
1681 if (!child) {
1682 struct dentry *new;
1683 new = d_alloc(dir, &qname);
1684 if (new) {
1685 child = instantiate(dir->d_inode, new, task, ptr);
1686 if (child)
1687 dput(new);
1688 else
1689 child = new;
1690 }
1691 }
1692 if (!child || IS_ERR(child) || !child->d_inode)
1693 goto end_instantiate;
1694 inode = child->d_inode;
1695 if (inode) {
1696 ino = inode->i_ino;
1697 type = inode->i_mode >> 12;
1698 }
1699 dput(child);
1700end_instantiate:
1701 if (!ino)
1702 ino = find_inode_number(dir, &qname);
1703 if (!ino)
1704 ino = 1;
1705 return filldir(dirent, name, len, filp->f_pos, ino, type);
1706}
1707
1708static unsigned name_to_int(struct dentry *dentry)
1709{
1710 const char *name = dentry->d_name.name;
1711 int len = dentry->d_name.len;
1712 unsigned n = 0;
1713
1714 if (len > 1 && *name == '0')
1715 goto out;
1716 while (len-- > 0) {
1717 unsigned c = *name++ - '0';
1718 if (c > 9)
1719 goto out;
1720 if (n >= (~0U-9)/10)
1721 goto out;
1722 n *= 10;
1723 n += c;
1724 }
1725 return n;
1726out:
1727 return ~0U;
1728}
1729
1730#define PROC_FDINFO_MAX 64
1731
1732static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1733{
1734 struct task_struct *task = get_proc_task(inode);
1735 struct files_struct *files = NULL;
1736 struct file *file;
1737 int fd = proc_fd(inode);
1738
1739 if (task) {
1740 files = get_files_struct(task);
1741 put_task_struct(task);
1742 }
1743 if (files) {
1744 /*
1745 * We are not taking a ref to the file structure, so we must
1746 * hold ->file_lock.
1747 */
1748 spin_lock(&files->file_lock);
1749 file = fcheck_files(files, fd);
1750 if (file) {
1751 unsigned int f_flags;
1752 struct fdtable *fdt;
1753
1754 fdt = files_fdtable(files);
1755 f_flags = file->f_flags & ~O_CLOEXEC;
1756 if (close_on_exec(fd, fdt))
1757 f_flags |= O_CLOEXEC;
1758
1759 if (path) {
1760 *path = file->f_path;
1761 path_get(&file->f_path);
1762 }
1763 if (info)
1764 snprintf(info, PROC_FDINFO_MAX,
1765 "pos:\t%lli\n"
1766 "flags:\t0%o\n",
1767 (long long) file->f_pos,
1768 f_flags);
1769 spin_unlock(&files->file_lock);
1770 put_files_struct(files);
1771 return 0;
1772 }
1773 spin_unlock(&files->file_lock);
1774 put_files_struct(files);
1775 }
1776 return -ENOENT;
1777}
1778
1779static int proc_fd_link(struct dentry *dentry, struct path *path)
1780{
1781 return proc_fd_info(dentry->d_inode, path, NULL);
1782}
1783
1784static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1785{
1786 struct inode *inode;
1787 struct task_struct *task;
1788 int fd;
1789 struct files_struct *files;
1790 const struct cred *cred;
1791
1792 if (nd && nd->flags & LOOKUP_RCU)
1793 return -ECHILD;
1794
1795 inode = dentry->d_inode;
1796 task = get_proc_task(inode);
1797 fd = proc_fd(inode);
1798
1799 if (task) {
1800 files = get_files_struct(task);
1801 if (files) {
1802 struct file *file;
1803 rcu_read_lock();
1804 file = fcheck_files(files, fd);
1805 if (file) {
1806 unsigned f_mode = file->f_mode;
1807
1808 rcu_read_unlock();
1809 put_files_struct(files);
1810
1811 if (task_dumpable(task)) {
1812 rcu_read_lock();
1813 cred = __task_cred(task);
1814 inode->i_uid = cred->euid;
1815 inode->i_gid = cred->egid;
1816 rcu_read_unlock();
1817 } else {
1818 inode->i_uid = GLOBAL_ROOT_UID;
1819 inode->i_gid = GLOBAL_ROOT_GID;
1820 }
1821
1822 if (S_ISLNK(inode->i_mode)) {
1823 unsigned i_mode = S_IFLNK;
1824 if (f_mode & FMODE_READ)
1825 i_mode |= S_IRUSR | S_IXUSR;
1826 if (f_mode & FMODE_WRITE)
1827 i_mode |= S_IWUSR | S_IXUSR;
1828 inode->i_mode = i_mode;
1829 }
1830
1831 security_task_to_inode(task, inode);
1832 put_task_struct(task);
1833 return 1;
1834 }
1835 rcu_read_unlock();
1836 put_files_struct(files);
1837 }
1838 put_task_struct(task);
1839 }
1840 d_drop(dentry);
1841 return 0;
1842}
1843
1844static const struct dentry_operations tid_fd_dentry_operations =
1845{
1846 .d_revalidate = tid_fd_revalidate,
1847 .d_delete = pid_delete_dentry,
1848};
1849
1850static struct dentry *proc_fd_instantiate(struct inode *dir,
1851 struct dentry *dentry, struct task_struct *task, const void *ptr)
1852{
1853 unsigned fd = (unsigned long)ptr;
1854 struct inode *inode;
1855 struct proc_inode *ei;
1856 struct dentry *error = ERR_PTR(-ENOENT);
1857
1858 inode = proc_pid_make_inode(dir->i_sb, task);
1859 if (!inode)
1860 goto out;
1861 ei = PROC_I(inode);
1862 ei->fd = fd;
1863
1864 inode->i_mode = S_IFLNK;
1865 inode->i_op = &proc_pid_link_inode_operations;
1866 inode->i_size = 64;
1867 ei->op.proc_get_link = proc_fd_link;
1868 d_set_d_op(dentry, &tid_fd_dentry_operations);
1869 d_add(dentry, inode);
1870 /* Close the race of the process dying before we return the dentry */
1871 if (tid_fd_revalidate(dentry, NULL))
1872 error = NULL;
1873
1874 out:
1875 return error;
1876}
1877
1878static struct dentry *proc_lookupfd_common(struct inode *dir,
1879 struct dentry *dentry,
1880 instantiate_t instantiate)
1881{
1882 struct task_struct *task = get_proc_task(dir);
1883 unsigned fd = name_to_int(dentry);
1884 struct dentry *result = ERR_PTR(-ENOENT);
1885
1886 if (!task)
1887 goto out_no_task;
1888 if (fd == ~0U)
1889 goto out;
1890
1891 result = instantiate(dir, dentry, task, (void *)(unsigned long)fd);
1892out:
1893 put_task_struct(task);
1894out_no_task:
1895 return result;
1896}
1897
1898static int proc_readfd_common(struct file * filp, void * dirent,
1899 filldir_t filldir, instantiate_t instantiate)
1900{
1901 struct dentry *dentry = filp->f_path.dentry;
1902 struct inode *inode = dentry->d_inode;
1903 struct task_struct *p = get_proc_task(inode);
1904 unsigned int fd, ino;
1905 int retval;
1906 struct files_struct * files;
1907
1908 retval = -ENOENT;
1909 if (!p)
1910 goto out_no_task;
1911 retval = 0;
1912
1913 fd = filp->f_pos;
1914 switch (fd) {
1915 case 0:
1916 if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1917 goto out;
1918 filp->f_pos++;
1919 case 1:
1920 ino = parent_ino(dentry);
1921 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1922 goto out;
1923 filp->f_pos++;
1924 default:
1925 files = get_files_struct(p);
1926 if (!files)
1927 goto out;
1928 rcu_read_lock();
1929 for (fd = filp->f_pos-2;
1930 fd < files_fdtable(files)->max_fds;
1931 fd++, filp->f_pos++) {
1932 char name[PROC_NUMBUF];
1933 int len;
1934 int rv;
1935
1936 if (!fcheck_files(files, fd))
1937 continue;
1938 rcu_read_unlock();
1939
1940 len = snprintf(name, sizeof(name), "%d", fd);
1941 rv = proc_fill_cache(filp, dirent, filldir,
1942 name, len, instantiate, p,
1943 (void *)(unsigned long)fd);
1944 if (rv < 0)
1945 goto out_fd_loop;
1946 rcu_read_lock();
1947 }
1948 rcu_read_unlock();
1949out_fd_loop:
1950 put_files_struct(files);
1951 }
1952out:
1953 put_task_struct(p);
1954out_no_task:
1955 return retval;
1956}
1957
1958static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1959 struct nameidata *nd)
1960{
1961 return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1962}
1963
1964static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1965{
1966 return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1967}
1968
1969static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1970 size_t len, loff_t *ppos)
1971{
1972 char tmp[PROC_FDINFO_MAX];
1973 int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1974 if (!err)
1975 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1976 return err;
1977}
1978
1979static const struct file_operations proc_fdinfo_file_operations = {
1980 .open = nonseekable_open,
1981 .read = proc_fdinfo_read,
1982 .llseek = no_llseek,
1983};
1984
1985static const struct file_operations proc_fd_operations = {
1986 .read = generic_read_dir,
1987 .readdir = proc_readfd,
1988 .llseek = default_llseek,
1989};
1990
1991#ifdef CONFIG_CHECKPOINT_RESTORE
1992
1993/*
1994 * dname_to_vma_addr - maps a dentry name into two unsigned longs
1995 * which represent vma start and end addresses.
1996 */
1997static int dname_to_vma_addr(struct dentry *dentry,
1998 unsigned long *start, unsigned long *end)
1999{
2000 if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
2001 return -EINVAL;
2002
2003 return 0;
2004}
2005
2006static int map_files_d_revalidate(struct dentry *dentry, struct nameidata *nd)
2007{
2008 unsigned long vm_start, vm_end;
2009 bool exact_vma_exists = false;
2010 struct mm_struct *mm = NULL;
2011 struct task_struct *task;
2012 const struct cred *cred;
2013 struct inode *inode;
2014 int status = 0;
2015
2016 if (nd && nd->flags & LOOKUP_RCU)
2017 return -ECHILD;
2018
2019 if (!capable(CAP_SYS_ADMIN)) {
2020 status = -EACCES;
2021 goto out_notask;
2022 }
2023
2024 inode = dentry->d_inode;
2025 task = get_proc_task(inode);
2026 if (!task)
2027 goto out_notask;
2028
2029 mm = mm_access(task, PTRACE_MODE_READ);
2030 if (IS_ERR_OR_NULL(mm))
2031 goto out;
2032
2033 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2034 down_read(&mm->mmap_sem);
2035 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
2036 up_read(&mm->mmap_sem);
2037 }
2038
2039 mmput(mm);
2040
2041 if (exact_vma_exists) {
2042 if (task_dumpable(task)) {
2043 rcu_read_lock();
2044 cred = __task_cred(task);
2045 inode->i_uid = cred->euid;
2046 inode->i_gid = cred->egid;
2047 rcu_read_unlock();
2048 } else {
2049 inode->i_uid = GLOBAL_ROOT_UID;
2050 inode->i_gid = GLOBAL_ROOT_GID;
2051 }
2052 security_task_to_inode(task, inode);
2053 status = 1;
2054 }
2055
2056out:
2057 put_task_struct(task);
2058
2059out_notask:
2060 if (status <= 0)
2061 d_drop(dentry);
2062
2063 return status;
2064}
2065
2066static const struct dentry_operations tid_map_files_dentry_operations = {
2067 .d_revalidate = map_files_d_revalidate,
2068 .d_delete = pid_delete_dentry,
2069};
2070
2071static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
2072{
2073 unsigned long vm_start, vm_end;
2074 struct vm_area_struct *vma;
2075 struct task_struct *task;
2076 struct mm_struct *mm;
2077 int rc;
2078
2079 rc = -ENOENT;
2080 task = get_proc_task(dentry->d_inode);
2081 if (!task)
2082 goto out;
2083
2084 mm = get_task_mm(task);
2085 put_task_struct(task);
2086 if (!mm)
2087 goto out;
2088
2089 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2090 if (rc)
2091 goto out_mmput;
2092
2093 down_read(&mm->mmap_sem);
2094 vma = find_exact_vma(mm, vm_start, vm_end);
2095 if (vma && vma->vm_file) {
2096 *path = vma->vm_file->f_path;
2097 path_get(path);
2098 rc = 0;
2099 }
2100 up_read(&mm->mmap_sem);
2101
2102out_mmput:
2103 mmput(mm);
2104out:
2105 return rc;
2106}
2107
2108struct map_files_info {
2109 struct file *file;
2110 unsigned long len;
2111 unsigned char name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2112};
2113
2114static struct dentry *
2115proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2116 struct task_struct *task, const void *ptr)
2117{
2118 const struct file *file = ptr;
2119 struct proc_inode *ei;
2120 struct inode *inode;
2121
2122 if (!file)
2123 return ERR_PTR(-ENOENT);
2124
2125 inode = proc_pid_make_inode(dir->i_sb, task);
2126 if (!inode)
2127 return ERR_PTR(-ENOENT);
2128
2129 ei = PROC_I(inode);
2130 ei->op.proc_get_link = proc_map_files_get_link;
2131
2132 inode->i_op = &proc_pid_link_inode_operations;
2133 inode->i_size = 64;
2134 inode->i_mode = S_IFLNK;
2135
2136 if (file->f_mode & FMODE_READ)
2137 inode->i_mode |= S_IRUSR;
2138 if (file->f_mode & FMODE_WRITE)
2139 inode->i_mode |= S_IWUSR;
2140
2141 d_set_d_op(dentry, &tid_map_files_dentry_operations);
2142 d_add(dentry, inode);
2143
2144 return NULL;
2145}
2146
2147static struct dentry *proc_map_files_lookup(struct inode *dir,
2148 struct dentry *dentry, struct nameidata *nd)
2149{
2150 unsigned long vm_start, vm_end;
2151 struct vm_area_struct *vma;
2152 struct task_struct *task;
2153 struct dentry *result;
2154 struct mm_struct *mm;
2155
2156 result = ERR_PTR(-EACCES);
2157 if (!capable(CAP_SYS_ADMIN))
2158 goto out;
2159
2160 result = ERR_PTR(-ENOENT);
2161 task = get_proc_task(dir);
2162 if (!task)
2163 goto out;
2164
2165 result = ERR_PTR(-EACCES);
2166 if (!ptrace_may_access(task, PTRACE_MODE_READ))
2167 goto out_put_task;
2168
2169 result = ERR_PTR(-ENOENT);
2170 if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2171 goto out_put_task;
2172
2173 mm = get_task_mm(task);
2174 if (!mm)
2175 goto out_put_task;
2176
2177 down_read(&mm->mmap_sem);
2178 vma = find_exact_vma(mm, vm_start, vm_end);
2179 if (!vma)
2180 goto out_no_vma;
2181
2182 result = proc_map_files_instantiate(dir, dentry, task, vma->vm_file);
2183
2184out_no_vma:
2185 up_read(&mm->mmap_sem);
2186 mmput(mm);
2187out_put_task:
2188 put_task_struct(task);
2189out:
2190 return result;
2191}
2192
2193static const struct inode_operations proc_map_files_inode_operations = {
2194 .lookup = proc_map_files_lookup,
2195 .permission = proc_fd_permission,
2196 .setattr = proc_setattr,
2197};
2198
2199static int
2200proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
2201{
2202 struct dentry *dentry = filp->f_path.dentry;
2203 struct inode *inode = dentry->d_inode;
2204 struct vm_area_struct *vma;
2205 struct task_struct *task;
2206 struct mm_struct *mm;
2207 ino_t ino;
2208 int ret;
2209
2210 ret = -EACCES;
2211 if (!capable(CAP_SYS_ADMIN))
2212 goto out;
2213
2214 ret = -ENOENT;
2215 task = get_proc_task(inode);
2216 if (!task)
2217 goto out;
2218
2219 ret = -EACCES;
2220 if (!ptrace_may_access(task, PTRACE_MODE_READ))
2221 goto out_put_task;
2222
2223 ret = 0;
2224 switch (filp->f_pos) {
2225 case 0:
2226 ino = inode->i_ino;
2227 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
2228 goto out_put_task;
2229 filp->f_pos++;
2230 case 1:
2231 ino = parent_ino(dentry);
2232 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2233 goto out_put_task;
2234 filp->f_pos++;
2235 default:
2236 {
2237 unsigned long nr_files, pos, i;
2238 struct flex_array *fa = NULL;
2239 struct map_files_info info;
2240 struct map_files_info *p;
2241
2242 mm = get_task_mm(task);
2243 if (!mm)
2244 goto out_put_task;
2245 down_read(&mm->mmap_sem);
2246
2247 nr_files = 0;
2248
2249 /*
2250 * We need two passes here:
2251 *
2252 * 1) Collect vmas of mapped files with mmap_sem taken
2253 * 2) Release mmap_sem and instantiate entries
2254 *
2255 * otherwise we get lockdep complained, since filldir()
2256 * routine might require mmap_sem taken in might_fault().
2257 */
2258
2259 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2260 if (vma->vm_file && ++pos > filp->f_pos)
2261 nr_files++;
2262 }
2263
2264 if (nr_files) {
2265 fa = flex_array_alloc(sizeof(info), nr_files,
2266 GFP_KERNEL);
2267 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2268 GFP_KERNEL)) {
2269 ret = -ENOMEM;
2270 if (fa)
2271 flex_array_free(fa);
2272 up_read(&mm->mmap_sem);
2273 mmput(mm);
2274 goto out_put_task;
2275 }
2276 for (i = 0, vma = mm->mmap, pos = 2; vma;
2277 vma = vma->vm_next) {
2278 if (!vma->vm_file)
2279 continue;
2280 if (++pos <= filp->f_pos)
2281 continue;
2282
2283 get_file(vma->vm_file);
2284 info.file = vma->vm_file;
2285 info.len = snprintf(info.name,
2286 sizeof(info.name), "%lx-%lx",
2287 vma->vm_start, vma->vm_end);
2288 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2289 BUG();
2290 }
2291 }
2292 up_read(&mm->mmap_sem);
2293
2294 for (i = 0; i < nr_files; i++) {
2295 p = flex_array_get(fa, i);
2296 ret = proc_fill_cache(filp, dirent, filldir,
2297 p->name, p->len,
2298 proc_map_files_instantiate,
2299 task, p->file);
2300 if (ret)
2301 break;
2302 filp->f_pos++;
2303 fput(p->file);
2304 }
2305 for (; i < nr_files; i++) {
2306 /*
2307 * In case of error don't forget
2308 * to put rest of file refs.
2309 */
2310 p = flex_array_get(fa, i);
2311 fput(p->file);
2312 }
2313 if (fa)
2314 flex_array_free(fa);
2315 mmput(mm);
2316 }
2317 }
2318
2319out_put_task:
2320 put_task_struct(task);
2321out:
2322 return ret;
2323}
2324
2325static const struct file_operations proc_map_files_operations = {
2326 .read = generic_read_dir,
2327 .readdir = proc_map_files_readdir,
2328 .llseek = default_llseek,
2329};
2330
2331#endif /* CONFIG_CHECKPOINT_RESTORE */
2332
2333/*
2334 * /proc/pid/fd needs a special permission handler so that a process can still
2335 * access /proc/self/fd after it has executed a setuid().
2336 */
2337static int proc_fd_permission(struct inode *inode, int mask)
2338{
2339 int rv = generic_permission(inode, mask);
2340 if (rv == 0)
2341 return 0;
2342 if (task_pid(current) == proc_pid(inode))
2343 rv = 0;
2344 return rv;
2345}
2346
2347/*
2348 * proc directories can do almost nothing..
2349 */
2350static const struct inode_operations proc_fd_inode_operations = {
2351 .lookup = proc_lookupfd,
2352 .permission = proc_fd_permission,
2353 .setattr = proc_setattr,
2354};
2355
2356static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2357 struct dentry *dentry, struct task_struct *task, const void *ptr)
2358{
2359 unsigned fd = (unsigned long)ptr;
2360 struct inode *inode;
2361 struct proc_inode *ei;
2362 struct dentry *error = ERR_PTR(-ENOENT);
2363
2364 inode = proc_pid_make_inode(dir->i_sb, task);
2365 if (!inode)
2366 goto out;
2367 ei = PROC_I(inode);
2368 ei->fd = fd;
2369 inode->i_mode = S_IFREG | S_IRUSR;
2370 inode->i_fop = &proc_fdinfo_file_operations;
2371 d_set_d_op(dentry, &tid_fd_dentry_operations);
2372 d_add(dentry, inode);
2373 /* Close the race of the process dying before we return the dentry */
2374 if (tid_fd_revalidate(dentry, NULL))
2375 error = NULL;
2376
2377 out:
2378 return error;
2379}
2380
2381static struct dentry *proc_lookupfdinfo(struct inode *dir,
2382 struct dentry *dentry,
2383 struct nameidata *nd)
2384{
2385 return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2386}
2387
2388static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2389{
2390 return proc_readfd_common(filp, dirent, filldir,
2391 proc_fdinfo_instantiate);
2392}
2393
2394static const struct file_operations proc_fdinfo_operations = {
2395 .read = generic_read_dir,
2396 .readdir = proc_readfdinfo,
2397 .llseek = default_llseek,
2398};
2399
2400/*
2401 * proc directories can do almost nothing..
2402 */
2403static const struct inode_operations proc_fdinfo_inode_operations = {
2404 .lookup = proc_lookupfdinfo,
2405 .setattr = proc_setattr,
2406};
2407
2408
2409static struct dentry *proc_pident_instantiate(struct inode *dir,
2410 struct dentry *dentry, struct task_struct *task, const void *ptr)
2411{
2412 const struct pid_entry *p = ptr;
2413 struct inode *inode;
2414 struct proc_inode *ei;
2415 struct dentry *error = ERR_PTR(-ENOENT);
2416
2417 inode = proc_pid_make_inode(dir->i_sb, task);
2418 if (!inode)
2419 goto out;
2420
2421 ei = PROC_I(inode);
2422 inode->i_mode = p->mode;
2423 if (S_ISDIR(inode->i_mode))
2424 set_nlink(inode, 2); /* Use getattr to fix if necessary */
2425 if (p->iop)
2426 inode->i_op = p->iop;
2427 if (p->fop)
2428 inode->i_fop = p->fop;
2429 ei->op = p->op;
2430 d_set_d_op(dentry, &pid_dentry_operations);
2431 d_add(dentry, inode);
2432 /* Close the race of the process dying before we return the dentry */
2433 if (pid_revalidate(dentry, NULL))
2434 error = NULL;
2435out:
2436 return error;
2437}
2438
2439static struct dentry *proc_pident_lookup(struct inode *dir,
2440 struct dentry *dentry,
2441 const struct pid_entry *ents,
2442 unsigned int nents)
2443{
2444 struct dentry *error;
2445 struct task_struct *task = get_proc_task(dir);
2446 const struct pid_entry *p, *last;
2447
2448 error = ERR_PTR(-ENOENT);
2449
2450 if (!task)
2451 goto out_no_task;
2452
2453 /*
2454 * Yes, it does not scale. And it should not. Don't add
2455 * new entries into /proc/<tgid>/ without very good reasons.
2456 */
2457 last = &ents[nents - 1];
2458 for (p = ents; p <= last; p++) {
2459 if (p->len != dentry->d_name.len)
2460 continue;
2461 if (!memcmp(dentry->d_name.name, p->name, p->len))
2462 break;
2463 }
2464 if (p > last)
2465 goto out;
2466
2467 error = proc_pident_instantiate(dir, dentry, task, p);
2468out:
2469 put_task_struct(task);
2470out_no_task:
2471 return error;
2472}
2473
2474static int proc_pident_fill_cache(struct file *filp, void *dirent,
2475 filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2476{
2477 return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2478 proc_pident_instantiate, task, p);
2479}
2480
2481static int proc_pident_readdir(struct file *filp,
2482 void *dirent, filldir_t filldir,
2483 const struct pid_entry *ents, unsigned int nents)
2484{
2485 int i;
2486 struct dentry *dentry = filp->f_path.dentry;
2487 struct inode *inode = dentry->d_inode;
2488 struct task_struct *task = get_proc_task(inode);
2489 const struct pid_entry *p, *last;
2490 ino_t ino;
2491 int ret;
2492
2493 ret = -ENOENT;
2494 if (!task)
2495 goto out_no_task;
2496
2497 ret = 0;
2498 i = filp->f_pos;
2499 switch (i) {
2500 case 0:
2501 ino = inode->i_ino;
2502 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2503 goto out;
2504 i++;
2505 filp->f_pos++;
2506 /* fall through */
2507 case 1:
2508 ino = parent_ino(dentry);
2509 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2510 goto out;
2511 i++;
2512 filp->f_pos++;
2513 /* fall through */
2514 default:
2515 i -= 2;
2516 if (i >= nents) {
2517 ret = 1;
2518 goto out;
2519 }
2520 p = ents + i;
2521 last = &ents[nents - 1];
2522 while (p <= last) {
2523 if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2524 goto out;
2525 filp->f_pos++;
2526 p++;
2527 }
2528 }
2529
2530 ret = 1;
2531out:
2532 put_task_struct(task);
2533out_no_task:
2534 return ret;
2535}
2536
2537#ifdef CONFIG_SECURITY
2538static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2539 size_t count, loff_t *ppos)
2540{
2541 struct inode * inode = file->f_path.dentry->d_inode;
2542 char *p = NULL;
2543 ssize_t length;
2544 struct task_struct *task = get_proc_task(inode);
2545
2546 if (!task)
2547 return -ESRCH;
2548
2549 length = security_getprocattr(task,
2550 (char*)file->f_path.dentry->d_name.name,
2551 &p);
2552 put_task_struct(task);
2553 if (length > 0)
2554 length = simple_read_from_buffer(buf, count, ppos, p, length);
2555 kfree(p);
2556 return length;
2557}
2558
2559static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2560 size_t count, loff_t *ppos)
2561{
2562 struct inode * inode = file->f_path.dentry->d_inode;
2563 char *page;
2564 ssize_t length;
2565 struct task_struct *task = get_proc_task(inode);
2566
2567 length = -ESRCH;
2568 if (!task)
2569 goto out_no_task;
2570 if (count > PAGE_SIZE)
2571 count = PAGE_SIZE;
2572
2573 /* No partial writes. */
2574 length = -EINVAL;
2575 if (*ppos != 0)
2576 goto out;
2577
2578 length = -ENOMEM;
2579 page = (char*)__get_free_page(GFP_TEMPORARY);
2580 if (!page)
2581 goto out;
2582
2583 length = -EFAULT;
2584 if (copy_from_user(page, buf, count))
2585 goto out_free;
2586
2587 /* Guard against adverse ptrace interaction */
2588 length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2589 if (length < 0)
2590 goto out_free;
2591
2592 length = security_setprocattr(task,
2593 (char*)file->f_path.dentry->d_name.name,
2594 (void*)page, count);
2595 mutex_unlock(&task->signal->cred_guard_mutex);
2596out_free:
2597 free_page((unsigned long) page);
2598out:
2599 put_task_struct(task);
2600out_no_task:
2601 return length;
2602}
2603
2604static const struct file_operations proc_pid_attr_operations = {
2605 .read = proc_pid_attr_read,
2606 .write = proc_pid_attr_write,
2607 .llseek = generic_file_llseek,
2608};
2609
2610static const struct pid_entry attr_dir_stuff[] = {
2611 REG("current", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2612 REG("prev", S_IRUGO, proc_pid_attr_operations),
2613 REG("exec", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2614 REG("fscreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2615 REG("keycreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2616 REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2617};
2618
2619static int proc_attr_dir_readdir(struct file * filp,
2620 void * dirent, filldir_t filldir)
2621{
2622 return proc_pident_readdir(filp,dirent,filldir,
2623 attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2624}
2625
2626static const struct file_operations proc_attr_dir_operations = {
2627 .read = generic_read_dir,
2628 .readdir = proc_attr_dir_readdir,
2629 .llseek = default_llseek,
2630};
2631
2632static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2633 struct dentry *dentry, struct nameidata *nd)
2634{
2635 return proc_pident_lookup(dir, dentry,
2636 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2637}
2638
2639static const struct inode_operations proc_attr_dir_inode_operations = {
2640 .lookup = proc_attr_dir_lookup,
2641 .getattr = pid_getattr,
2642 .setattr = proc_setattr,
2643};
2644
2645#endif
2646
2647#ifdef CONFIG_ELF_CORE
2648static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2649 size_t count, loff_t *ppos)
2650{
2651 struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2652 struct mm_struct *mm;
2653 char buffer[PROC_NUMBUF];
2654 size_t len;
2655 int ret;
2656
2657 if (!task)
2658 return -ESRCH;
2659
2660 ret = 0;
2661 mm = get_task_mm(task);
2662 if (mm) {
2663 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2664 ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2665 MMF_DUMP_FILTER_SHIFT));
2666 mmput(mm);
2667 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2668 }
2669
2670 put_task_struct(task);
2671
2672 return ret;
2673}
2674
2675static ssize_t proc_coredump_filter_write(struct file *file,
2676 const char __user *buf,
2677 size_t count,
2678 loff_t *ppos)
2679{
2680 struct task_struct *task;
2681 struct mm_struct *mm;
2682 char buffer[PROC_NUMBUF], *end;
2683 unsigned int val;
2684 int ret;
2685 int i;
2686 unsigned long mask;
2687
2688 ret = -EFAULT;
2689 memset(buffer, 0, sizeof(buffer));
2690 if (count > sizeof(buffer) - 1)
2691 count = sizeof(buffer) - 1;
2692 if (copy_from_user(buffer, buf, count))
2693 goto out_no_task;
2694
2695 ret = -EINVAL;
2696 val = (unsigned int)simple_strtoul(buffer, &end, 0);
2697 if (*end == '\n')
2698 end++;
2699 if (end - buffer == 0)
2700 goto out_no_task;
2701
2702 ret = -ESRCH;
2703 task = get_proc_task(file->f_dentry->d_inode);
2704 if (!task)
2705 goto out_no_task;
2706
2707 ret = end - buffer;
2708 mm = get_task_mm(task);
2709 if (!mm)
2710 goto out_no_mm;
2711
2712 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2713 if (val & mask)
2714 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2715 else
2716 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2717 }
2718
2719 mmput(mm);
2720 out_no_mm:
2721 put_task_struct(task);
2722 out_no_task:
2723 return ret;
2724}
2725
2726static const struct file_operations proc_coredump_filter_operations = {
2727 .read = proc_coredump_filter_read,
2728 .write = proc_coredump_filter_write,
2729 .llseek = generic_file_llseek,
2730};
2731#endif
2732
2733/*
2734 * /proc/self:
2735 */
2736static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2737 int buflen)
2738{
2739 struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2740 pid_t tgid = task_tgid_nr_ns(current, ns);
2741 char tmp[PROC_NUMBUF];
2742 if (!tgid)
2743 return -ENOENT;
2744 sprintf(tmp, "%d", tgid);
2745 return vfs_readlink(dentry,buffer,buflen,tmp);
2746}
2747
2748static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2749{
2750 struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2751 pid_t tgid = task_tgid_nr_ns(current, ns);
2752 char *name = ERR_PTR(-ENOENT);
2753 if (tgid) {
2754 name = __getname();
2755 if (!name)
2756 name = ERR_PTR(-ENOMEM);
2757 else
2758 sprintf(name, "%d", tgid);
2759 }
2760 nd_set_link(nd, name);
2761 return NULL;
2762}
2763
2764static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2765 void *cookie)
2766{
2767 char *s = nd_get_link(nd);
2768 if (!IS_ERR(s))
2769 __putname(s);
2770}
2771
2772static const struct inode_operations proc_self_inode_operations = {
2773 .readlink = proc_self_readlink,
2774 .follow_link = proc_self_follow_link,
2775 .put_link = proc_self_put_link,
2776};
2777
2778/*
2779 * proc base
2780 *
2781 * These are the directory entries in the root directory of /proc
2782 * that properly belong to the /proc filesystem, as they describe
2783 * describe something that is process related.
2784 */
2785static const struct pid_entry proc_base_stuff[] = {
2786 NOD("self", S_IFLNK|S_IRWXUGO,
2787 &proc_self_inode_operations, NULL, {}),
2788};
2789
2790static struct dentry *proc_base_instantiate(struct inode *dir,
2791 struct dentry *dentry, struct task_struct *task, const void *ptr)
2792{
2793 const struct pid_entry *p = ptr;
2794 struct inode *inode;
2795 struct proc_inode *ei;
2796 struct dentry *error;
2797
2798 /* Allocate the inode */
2799 error = ERR_PTR(-ENOMEM);
2800 inode = new_inode(dir->i_sb);
2801 if (!inode)
2802 goto out;
2803
2804 /* Initialize the inode */
2805 ei = PROC_I(inode);
2806 inode->i_ino = get_next_ino();
2807 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2808
2809 /*
2810 * grab the reference to the task.
2811 */
2812 ei->pid = get_task_pid(task, PIDTYPE_PID);
2813 if (!ei->pid)
2814 goto out_iput;
2815
2816 inode->i_mode = p->mode;
2817 if (S_ISDIR(inode->i_mode))
2818 set_nlink(inode, 2);
2819 if (S_ISLNK(inode->i_mode))
2820 inode->i_size = 64;
2821 if (p->iop)
2822 inode->i_op = p->iop;
2823 if (p->fop)
2824 inode->i_fop = p->fop;
2825 ei->op = p->op;
2826 d_add(dentry, inode);
2827 error = NULL;
2828out:
2829 return error;
2830out_iput:
2831 iput(inode);
2832 goto out;
2833}
2834
2835static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2836{
2837 struct dentry *error;
2838 struct task_struct *task = get_proc_task(dir);
2839 const struct pid_entry *p, *last;
2840
2841 error = ERR_PTR(-ENOENT);
2842
2843 if (!task)
2844 goto out_no_task;
2845
2846 /* Lookup the directory entry */
2847 last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2848 for (p = proc_base_stuff; p <= last; p++) {
2849 if (p->len != dentry->d_name.len)
2850 continue;
2851 if (!memcmp(dentry->d_name.name, p->name, p->len))
2852 break;
2853 }
2854 if (p > last)
2855 goto out;
2856
2857 error = proc_base_instantiate(dir, dentry, task, p);
2858
2859out:
2860 put_task_struct(task);
2861out_no_task:
2862 return error;
2863}
2864
2865static int proc_base_fill_cache(struct file *filp, void *dirent,
2866 filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2867{
2868 return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2869 proc_base_instantiate, task, p);
2870}
2871
2872#ifdef CONFIG_TASK_IO_ACCOUNTING
2873static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2874{
2875 struct task_io_accounting acct = task->ioac;
2876 unsigned long flags;
2877 int result;
2878
2879 result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2880 if (result)
2881 return result;
2882
2883 if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2884 result = -EACCES;
2885 goto out_unlock;
2886 }
2887
2888 if (whole && lock_task_sighand(task, &flags)) {
2889 struct task_struct *t = task;
2890
2891 task_io_accounting_add(&acct, &task->signal->ioac);
2892 while_each_thread(task, t)
2893 task_io_accounting_add(&acct, &t->ioac);
2894
2895 unlock_task_sighand(task, &flags);
2896 }
2897 result = sprintf(buffer,
2898 "rchar: %llu\n"
2899 "wchar: %llu\n"
2900 "syscr: %llu\n"
2901 "syscw: %llu\n"
2902 "read_bytes: %llu\n"
2903 "write_bytes: %llu\n"
2904 "cancelled_write_bytes: %llu\n",
2905 (unsigned long long)acct.rchar,
2906 (unsigned long long)acct.wchar,
2907 (unsigned long long)acct.syscr,
2908 (unsigned long long)acct.syscw,
2909 (unsigned long long)acct.read_bytes,
2910 (unsigned long long)acct.write_bytes,
2911 (unsigned long long)acct.cancelled_write_bytes);
2912out_unlock:
2913 mutex_unlock(&task->signal->cred_guard_mutex);
2914 return result;
2915}
2916
2917static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2918{
2919 return do_io_accounting(task, buffer, 0);
2920}
2921
2922static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2923{
2924 return do_io_accounting(task, buffer, 1);
2925}
2926#endif /* CONFIG_TASK_IO_ACCOUNTING */
2927
2928#ifdef CONFIG_USER_NS
2929static int proc_id_map_open(struct inode *inode, struct file *file,
2930 struct seq_operations *seq_ops)
2931{
2932 struct user_namespace *ns = NULL;
2933 struct task_struct *task;
2934 struct seq_file *seq;
2935 int ret = -EINVAL;
2936
2937 task = get_proc_task(inode);
2938 if (task) {
2939 rcu_read_lock();
2940 ns = get_user_ns(task_cred_xxx(task, user_ns));
2941 rcu_read_unlock();
2942 put_task_struct(task);
2943 }
2944 if (!ns)
2945 goto err;
2946
2947 ret = seq_open(file, seq_ops);
2948 if (ret)
2949 goto err_put_ns;
2950
2951 seq = file->private_data;
2952 seq->private = ns;
2953
2954 return 0;
2955err_put_ns:
2956 put_user_ns(ns);
2957err:
2958 return ret;
2959}
2960
2961static int proc_id_map_release(struct inode *inode, struct file *file)
2962{
2963 struct seq_file *seq = file->private_data;
2964 struct user_namespace *ns = seq->private;
2965 put_user_ns(ns);
2966 return seq_release(inode, file);
2967}
2968
2969static int proc_uid_map_open(struct inode *inode, struct file *file)
2970{
2971 return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2972}
2973
2974static int proc_gid_map_open(struct inode *inode, struct file *file)
2975{
2976 return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2977}
2978
2979static const struct file_operations proc_uid_map_operations = {
2980 .open = proc_uid_map_open,
2981 .write = proc_uid_map_write,
2982 .read = seq_read,
2983 .llseek = seq_lseek,
2984 .release = proc_id_map_release,
2985};
2986
2987static const struct file_operations proc_gid_map_operations = {
2988 .open = proc_gid_map_open,
2989 .write = proc_gid_map_write,
2990 .read = seq_read,
2991 .llseek = seq_lseek,
2992 .release = proc_id_map_release,
2993};
2994#endif /* CONFIG_USER_NS */
2995
2996static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2997 struct pid *pid, struct task_struct *task)
2998{
2999 int err = lock_trace(task);
3000 if (!err) {
3001 seq_printf(m, "%08x\n", task->personality);
3002 unlock_trace(task);
3003 }
3004 return err;
3005}
3006
3007/*
3008 * Thread groups
3009 */
3010static const struct file_operations proc_task_operations;
3011static const struct inode_operations proc_task_inode_operations;
3012
3013static const struct pid_entry tgid_base_stuff[] = {
3014 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3015 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3016#ifdef CONFIG_CHECKPOINT_RESTORE
3017 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3018#endif
3019 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3020 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3021#ifdef CONFIG_NET
3022 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3023#endif
3024 REG("environ", S_IRUSR, proc_environ_operations),
3025 INF("auxv", S_IRUSR, proc_pid_auxv),
3026 ONE("status", S_IRUGO, proc_pid_status),
3027 ONE("personality", S_IRUGO, proc_pid_personality),
3028 INF("limits", S_IRUGO, proc_pid_limits),
3029#ifdef CONFIG_SCHED_DEBUG
3030 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3031#endif
3032#ifdef CONFIG_SCHED_AUTOGROUP
3033 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3034#endif
3035 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3036#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3037 INF("syscall", S_IRUGO, proc_pid_syscall),
3038#endif
3039 INF("cmdline", S_IRUGO, proc_pid_cmdline),
3040 ONE("stat", S_IRUGO, proc_tgid_stat),
3041 ONE("statm", S_IRUGO, proc_pid_statm),
3042 REG("maps", S_IRUGO, proc_pid_maps_operations),
3043#ifdef CONFIG_NUMA
3044 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3045#endif
3046 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3047 LNK("cwd", proc_cwd_link),
3048 LNK("root", proc_root_link),
3049 LNK("exe", proc_exe_link),
3050 REG("mounts", S_IRUGO, proc_mounts_operations),
3051 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3052 REG("mountstats", S_IRUSR, proc_mountstats_operations),
3053#ifdef CONFIG_PROC_PAGE_MONITOR
3054 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3055 REG("smaps", S_IRUGO, proc_pid_smaps_operations),
3056 REG("pagemap", S_IRUGO, proc_pagemap_operations),
3057#endif
3058#ifdef CONFIG_SECURITY
3059 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3060#endif
3061#ifdef CONFIG_KALLSYMS
3062 INF("wchan", S_IRUGO, proc_pid_wchan),
3063#endif
3064#ifdef CONFIG_STACKTRACE
3065 ONE("stack", S_IRUGO, proc_pid_stack),
3066#endif
3067#ifdef CONFIG_SCHEDSTATS
3068 INF("schedstat", S_IRUGO, proc_pid_schedstat),
3069#endif
3070#ifdef CONFIG_LATENCYTOP
3071 REG("latency", S_IRUGO, proc_lstats_operations),
3072#endif
3073#ifdef CONFIG_PROC_PID_CPUSET
3074 REG("cpuset", S_IRUGO, proc_cpuset_operations),
3075#endif
3076#ifdef CONFIG_CGROUPS
3077 REG("cgroup", S_IRUGO, proc_cgroup_operations),
3078#endif
3079 INF("oom_score", S_IRUGO, proc_oom_score),
3080 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3081 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3082#ifdef CONFIG_AUDITSYSCALL
3083 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3084 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3085#endif
3086#ifdef CONFIG_FAULT_INJECTION
3087 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3088#endif
3089#ifdef CONFIG_ELF_CORE
3090 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3091#endif
3092#ifdef CONFIG_TASK_IO_ACCOUNTING
3093 INF("io", S_IRUSR, proc_tgid_io_accounting),
3094#endif
3095#ifdef CONFIG_HARDWALL
3096 INF("hardwall", S_IRUGO, proc_pid_hardwall),
3097#endif
3098#ifdef CONFIG_USER_NS
3099 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3100 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3101#endif
3102};
3103
3104static int proc_tgid_base_readdir(struct file * filp,
3105 void * dirent, filldir_t filldir)
3106{
3107 return proc_pident_readdir(filp,dirent,filldir,
3108 tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
3109}
3110
3111static const struct file_operations proc_tgid_base_operations = {
3112 .read = generic_read_dir,
3113 .readdir = proc_tgid_base_readdir,
3114 .llseek = default_llseek,
3115};
3116
3117static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3118 return proc_pident_lookup(dir, dentry,
3119 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3120}
3121
3122static const struct inode_operations proc_tgid_base_inode_operations = {
3123 .lookup = proc_tgid_base_lookup,
3124 .getattr = pid_getattr,
3125 .setattr = proc_setattr,
3126 .permission = proc_pid_permission,
3127};
3128
3129static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3130{
3131 struct dentry *dentry, *leader, *dir;
3132 char buf[PROC_NUMBUF];
3133 struct qstr name;
3134
3135 name.name = buf;
3136 name.len = snprintf(buf, sizeof(buf), "%d", pid);
3137 dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3138 if (dentry) {
3139 shrink_dcache_parent(dentry);
3140 d_drop(dentry);
3141 dput(dentry);
3142 }
3143
3144 name.name = buf;
3145 name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3146 leader = d_hash_and_lookup(mnt->mnt_root, &name);
3147 if (!leader)
3148 goto out;
3149
3150 name.name = "task";
3151 name.len = strlen(name.name);
3152 dir = d_hash_and_lookup(leader, &name);
3153 if (!dir)
3154 goto out_put_leader;
3155
3156 name.name = buf;
3157 name.len = snprintf(buf, sizeof(buf), "%d", pid);
3158 dentry = d_hash_and_lookup(dir, &name);
3159 if (dentry) {
3160 shrink_dcache_parent(dentry);
3161 d_drop(dentry);
3162 dput(dentry);
3163 }
3164
3165 dput(dir);
3166out_put_leader:
3167 dput(leader);
3168out:
3169 return;
3170}
3171
3172/**
3173 * proc_flush_task - Remove dcache entries for @task from the /proc dcache.
3174 * @task: task that should be flushed.
3175 *
3176 * When flushing dentries from proc, one needs to flush them from global
3177 * proc (proc_mnt) and from all the namespaces' procs this task was seen
3178 * in. This call is supposed to do all of this job.
3179 *
3180 * Looks in the dcache for
3181 * /proc/@pid
3182 * /proc/@tgid/task/@pid
3183 * if either directory is present flushes it and all of it'ts children
3184 * from the dcache.
3185 *
3186 * It is safe and reasonable to cache /proc entries for a task until
3187 * that task exits. After that they just clog up the dcache with
3188 * useless entries, possibly causing useful dcache entries to be
3189 * flushed instead. This routine is proved to flush those useless
3190 * dcache entries at process exit time.
3191 *
3192 * NOTE: This routine is just an optimization so it does not guarantee
3193 * that no dcache entries will exist at process exit time it
3194 * just makes it very unlikely that any will persist.
3195 */
3196
3197void proc_flush_task(struct task_struct *task)
3198{
3199 int i;
3200 struct pid *pid, *tgid;
3201 struct upid *upid;
3202
3203 pid = task_pid(task);
3204 tgid = task_tgid(task);
3205
3206 for (i = 0; i <= pid->level; i++) {
3207 upid = &pid->numbers[i];
3208 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3209 tgid->numbers[i].nr);
3210 }
3211
3212 upid = &pid->numbers[pid->level];
3213 if (upid->nr == 1)
3214 pid_ns_release_proc(upid->ns);
3215}
3216
3217static struct dentry *proc_pid_instantiate(struct inode *dir,
3218 struct dentry * dentry,
3219 struct task_struct *task, const void *ptr)
3220{
3221 struct dentry *error = ERR_PTR(-ENOENT);
3222 struct inode *inode;
3223
3224 inode = proc_pid_make_inode(dir->i_sb, task);
3225 if (!inode)
3226 goto out;
3227
3228 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3229 inode->i_op = &proc_tgid_base_inode_operations;
3230 inode->i_fop = &proc_tgid_base_operations;
3231 inode->i_flags|=S_IMMUTABLE;
3232
3233 set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3234 ARRAY_SIZE(tgid_base_stuff)));
3235
3236 d_set_d_op(dentry, &pid_dentry_operations);
3237
3238 d_add(dentry, inode);
3239 /* Close the race of the process dying before we return the dentry */
3240 if (pid_revalidate(dentry, NULL))
3241 error = NULL;
3242out:
3243 return error;
3244}
3245
3246struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3247{
3248 struct dentry *result;
3249 struct task_struct *task;
3250 unsigned tgid;
3251 struct pid_namespace *ns;
3252
3253 result = proc_base_lookup(dir, dentry);
3254 if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3255 goto out;
3256
3257 tgid = name_to_int(dentry);
3258 if (tgid == ~0U)
3259 goto out;
3260
3261 ns = dentry->d_sb->s_fs_info;
3262 rcu_read_lock();
3263 task = find_task_by_pid_ns(tgid, ns);
3264 if (task)
3265 get_task_struct(task);
3266 rcu_read_unlock();
3267 if (!task)
3268 goto out;
3269
3270 result = proc_pid_instantiate(dir, dentry, task, NULL);
3271 put_task_struct(task);
3272out:
3273 return result;
3274}
3275
3276/*
3277 * Find the first task with tgid >= tgid
3278 *
3279 */
3280struct tgid_iter {
3281 unsigned int tgid;
3282 struct task_struct *task;
3283};
3284static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3285{
3286 struct pid *pid;
3287
3288 if (iter.task)
3289 put_task_struct(iter.task);
3290 rcu_read_lock();
3291retry:
3292 iter.task = NULL;
3293 pid = find_ge_pid(iter.tgid, ns);
3294 if (pid) {
3295 iter.tgid = pid_nr_ns(pid, ns);
3296 iter.task = pid_task(pid, PIDTYPE_PID);
3297 /* What we to know is if the pid we have find is the
3298 * pid of a thread_group_leader. Testing for task
3299 * being a thread_group_leader is the obvious thing
3300 * todo but there is a window when it fails, due to
3301 * the pid transfer logic in de_thread.
3302 *
3303 * So we perform the straight forward test of seeing
3304 * if the pid we have found is the pid of a thread
3305 * group leader, and don't worry if the task we have
3306 * found doesn't happen to be a thread group leader.
3307 * As we don't care in the case of readdir.
3308 */
3309 if (!iter.task || !has_group_leader_pid(iter.task)) {
3310 iter.tgid += 1;
3311 goto retry;
3312 }
3313 get_task_struct(iter.task);
3314 }
3315 rcu_read_unlock();
3316 return iter;
3317}
3318
3319#define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3320
3321static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3322 struct tgid_iter iter)
3323{
3324 char name[PROC_NUMBUF];
3325 int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3326 return proc_fill_cache(filp, dirent, filldir, name, len,
3327 proc_pid_instantiate, iter.task, NULL);
3328}
3329
3330static int fake_filldir(void *buf, const char *name, int namelen,
3331 loff_t offset, u64 ino, unsigned d_type)
3332{
3333 return 0;
3334}
3335
3336/* for the /proc/ directory itself, after non-process stuff has been done */
3337int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3338{
3339 unsigned int nr;
3340 struct task_struct *reaper;
3341 struct tgid_iter iter;
3342 struct pid_namespace *ns;
3343 filldir_t __filldir;
3344
3345 if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3346 goto out_no_task;
3347 nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3348
3349 reaper = get_proc_task(filp->f_path.dentry->d_inode);
3350 if (!reaper)
3351 goto out_no_task;
3352
3353 for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3354 const struct pid_entry *p = &proc_base_stuff[nr];
3355 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3356 goto out;
3357 }
3358
3359 ns = filp->f_dentry->d_sb->s_fs_info;
3360 iter.task = NULL;
3361 iter.tgid = filp->f_pos - TGID_OFFSET;
3362 for (iter = next_tgid(ns, iter);
3363 iter.task;
3364 iter.tgid += 1, iter = next_tgid(ns, iter)) {
3365 if (has_pid_permissions(ns, iter.task, 2))
3366 __filldir = filldir;
3367 else
3368 __filldir = fake_filldir;
3369
3370 filp->f_pos = iter.tgid + TGID_OFFSET;
3371 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3372 put_task_struct(iter.task);
3373 goto out;
3374 }
3375 }
3376 filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3377out:
3378 put_task_struct(reaper);
3379out_no_task:
3380 return 0;
3381}
3382
3383/*
3384 * Tasks
3385 */
3386static const struct pid_entry tid_base_stuff[] = {
3387 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3388 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3389 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3390 REG("environ", S_IRUSR, proc_environ_operations),
3391 INF("auxv", S_IRUSR, proc_pid_auxv),
3392 ONE("status", S_IRUGO, proc_pid_status),
3393 ONE("personality", S_IRUGO, proc_pid_personality),
3394 INF("limits", S_IRUGO, proc_pid_limits),
3395#ifdef CONFIG_SCHED_DEBUG
3396 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3397#endif
3398 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3399#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3400 INF("syscall", S_IRUGO, proc_pid_syscall),
3401#endif
3402 INF("cmdline", S_IRUGO, proc_pid_cmdline),
3403 ONE("stat", S_IRUGO, proc_tid_stat),
3404 ONE("statm", S_IRUGO, proc_pid_statm),
3405 REG("maps", S_IRUGO, proc_tid_maps_operations),
3406#ifdef CONFIG_CHECKPOINT_RESTORE
3407 REG("children", S_IRUGO, proc_tid_children_operations),
3408#endif
3409#ifdef CONFIG_NUMA
3410 REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3411#endif
3412 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations),
3413 LNK("cwd", proc_cwd_link),
3414 LNK("root", proc_root_link),
3415 LNK("exe", proc_exe_link),
3416 REG("mounts", S_IRUGO, proc_mounts_operations),
3417 REG("mountinfo", S_IRUGO, proc_mountinfo_operations),
3418#ifdef CONFIG_PROC_PAGE_MONITOR
3419 REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3420 REG("smaps", S_IRUGO, proc_tid_smaps_operations),
3421 REG("pagemap", S_IRUGO, proc_pagemap_operations),
3422#endif
3423#ifdef CONFIG_SECURITY
3424 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3425#endif
3426#ifdef CONFIG_KALLSYMS
3427 INF("wchan", S_IRUGO, proc_pid_wchan),
3428#endif
3429#ifdef CONFIG_STACKTRACE
3430 ONE("stack", S_IRUGO, proc_pid_stack),
3431#endif
3432#ifdef CONFIG_SCHEDSTATS
3433 INF("schedstat", S_IRUGO, proc_pid_schedstat),
3434#endif
3435#ifdef CONFIG_LATENCYTOP
3436 REG("latency", S_IRUGO, proc_lstats_operations),
3437#endif
3438#ifdef CONFIG_PROC_PID_CPUSET
3439 REG("cpuset", S_IRUGO, proc_cpuset_operations),
3440#endif
3441#ifdef CONFIG_CGROUPS
3442 REG("cgroup", S_IRUGO, proc_cgroup_operations),
3443#endif
3444 INF("oom_score", S_IRUGO, proc_oom_score),
3445 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3446 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3447#ifdef CONFIG_AUDITSYSCALL
3448 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations),
3449 REG("sessionid", S_IRUGO, proc_sessionid_operations),
3450#endif
3451#ifdef CONFIG_FAULT_INJECTION
3452 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3453#endif
3454#ifdef CONFIG_TASK_IO_ACCOUNTING
3455 INF("io", S_IRUSR, proc_tid_io_accounting),
3456#endif
3457#ifdef CONFIG_HARDWALL
3458 INF("hardwall", S_IRUGO, proc_pid_hardwall),
3459#endif
3460#ifdef CONFIG_USER_NS
3461 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations),
3462 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations),
3463#endif
3464};
3465
3466static int proc_tid_base_readdir(struct file * filp,
3467 void * dirent, filldir_t filldir)
3468{
3469 return proc_pident_readdir(filp,dirent,filldir,
3470 tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3471}
3472
3473static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3474 return proc_pident_lookup(dir, dentry,
3475 tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3476}
3477
3478static const struct file_operations proc_tid_base_operations = {
3479 .read = generic_read_dir,
3480 .readdir = proc_tid_base_readdir,
3481 .llseek = default_llseek,
3482};
3483
3484static const struct inode_operations proc_tid_base_inode_operations = {
3485 .lookup = proc_tid_base_lookup,
3486 .getattr = pid_getattr,
3487 .setattr = proc_setattr,
3488};
3489
3490static struct dentry *proc_task_instantiate(struct inode *dir,
3491 struct dentry *dentry, struct task_struct *task, const void *ptr)
3492{
3493 struct dentry *error = ERR_PTR(-ENOENT);
3494 struct inode *inode;
3495 inode = proc_pid_make_inode(dir->i_sb, task);
3496
3497 if (!inode)
3498 goto out;
3499 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3500 inode->i_op = &proc_tid_base_inode_operations;
3501 inode->i_fop = &proc_tid_base_operations;
3502 inode->i_flags|=S_IMMUTABLE;
3503
3504 set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3505 ARRAY_SIZE(tid_base_stuff)));
3506
3507 d_set_d_op(dentry, &pid_dentry_operations);
3508
3509 d_add(dentry, inode);
3510 /* Close the race of the process dying before we return the dentry */
3511 if (pid_revalidate(dentry, NULL))
3512 error = NULL;
3513out:
3514 return error;
3515}
3516
3517static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3518{
3519 struct dentry *result = ERR_PTR(-ENOENT);
3520 struct task_struct *task;
3521 struct task_struct *leader = get_proc_task(dir);
3522 unsigned tid;
3523 struct pid_namespace *ns;
3524
3525 if (!leader)
3526 goto out_no_task;
3527
3528 tid = name_to_int(dentry);
3529 if (tid == ~0U)
3530 goto out;
3531
3532 ns = dentry->d_sb->s_fs_info;
3533 rcu_read_lock();
3534 task = find_task_by_pid_ns(tid, ns);
3535 if (task)
3536 get_task_struct(task);
3537 rcu_read_unlock();
3538 if (!task)
3539 goto out;
3540 if (!same_thread_group(leader, task))
3541 goto out_drop_task;
3542
3543 result = proc_task_instantiate(dir, dentry, task, NULL);
3544out_drop_task:
3545 put_task_struct(task);
3546out:
3547 put_task_struct(leader);
3548out_no_task:
3549 return result;
3550}
3551
3552/*
3553 * Find the first tid of a thread group to return to user space.
3554 *
3555 * Usually this is just the thread group leader, but if the users
3556 * buffer was too small or there was a seek into the middle of the
3557 * directory we have more work todo.
3558 *
3559 * In the case of a short read we start with find_task_by_pid.
3560 *
3561 * In the case of a seek we start with the leader and walk nr
3562 * threads past it.
3563 */
3564static struct task_struct *first_tid(struct task_struct *leader,
3565 int tid, int nr, struct pid_namespace *ns)
3566{
3567 struct task_struct *pos;
3568
3569 rcu_read_lock();
3570 /* Attempt to start with the pid of a thread */
3571 if (tid && (nr > 0)) {
3572 pos = find_task_by_pid_ns(tid, ns);
3573 if (pos && (pos->group_leader == leader))
3574 goto found;
3575 }
3576
3577 /* If nr exceeds the number of threads there is nothing todo */
3578 pos = NULL;
3579 if (nr && nr >= get_nr_threads(leader))
3580 goto out;
3581
3582 /* If we haven't found our starting place yet start
3583 * with the leader and walk nr threads forward.
3584 */
3585 for (pos = leader; nr > 0; --nr) {
3586 pos = next_thread(pos);
3587 if (pos == leader) {
3588 pos = NULL;
3589 goto out;
3590 }
3591 }
3592found:
3593 get_task_struct(pos);
3594out:
3595 rcu_read_unlock();
3596 return pos;
3597}
3598
3599/*
3600 * Find the next thread in the thread list.
3601 * Return NULL if there is an error or no next thread.
3602 *
3603 * The reference to the input task_struct is released.
3604 */
3605static struct task_struct *next_tid(struct task_struct *start)
3606{
3607 struct task_struct *pos = NULL;
3608 rcu_read_lock();
3609 if (pid_alive(start)) {
3610 pos = next_thread(start);
3611 if (thread_group_leader(pos))
3612 pos = NULL;
3613 else
3614 get_task_struct(pos);
3615 }
3616 rcu_read_unlock();
3617 put_task_struct(start);
3618 return pos;
3619}
3620
3621static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3622 struct task_struct *task, int tid)
3623{
3624 char name[PROC_NUMBUF];
3625 int len = snprintf(name, sizeof(name), "%d", tid);
3626 return proc_fill_cache(filp, dirent, filldir, name, len,
3627 proc_task_instantiate, task, NULL);
3628}
3629
3630/* for the /proc/TGID/task/ directories */
3631static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3632{
3633 struct dentry *dentry = filp->f_path.dentry;
3634 struct inode *inode = dentry->d_inode;
3635 struct task_struct *leader = NULL;
3636 struct task_struct *task;
3637 int retval = -ENOENT;
3638 ino_t ino;
3639 int tid;
3640 struct pid_namespace *ns;
3641
3642 task = get_proc_task(inode);
3643 if (!task)
3644 goto out_no_task;
3645 rcu_read_lock();
3646 if (pid_alive(task)) {
3647 leader = task->group_leader;
3648 get_task_struct(leader);
3649 }
3650 rcu_read_unlock();
3651 put_task_struct(task);
3652 if (!leader)
3653 goto out_no_task;
3654 retval = 0;
3655
3656 switch ((unsigned long)filp->f_pos) {
3657 case 0:
3658 ino = inode->i_ino;
3659 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3660 goto out;
3661 filp->f_pos++;
3662 /* fall through */
3663 case 1:
3664 ino = parent_ino(dentry);
3665 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3666 goto out;
3667 filp->f_pos++;
3668 /* fall through */
3669 }
3670
3671 /* f_version caches the tgid value that the last readdir call couldn't
3672 * return. lseek aka telldir automagically resets f_version to 0.
3673 */
3674 ns = filp->f_dentry->d_sb->s_fs_info;
3675 tid = (int)filp->f_version;
3676 filp->f_version = 0;
3677 for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3678 task;
3679 task = next_tid(task), filp->f_pos++) {
3680 tid = task_pid_nr_ns(task, ns);
3681 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3682 /* returning this tgid failed, save it as the first
3683 * pid for the next readir call */
3684 filp->f_version = (u64)tid;
3685 put_task_struct(task);
3686 break;
3687 }
3688 }
3689out:
3690 put_task_struct(leader);
3691out_no_task:
3692 return retval;
3693}
3694
3695static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3696{
3697 struct inode *inode = dentry->d_inode;
3698 struct task_struct *p = get_proc_task(inode);
3699 generic_fillattr(inode, stat);
3700
3701 if (p) {
3702 stat->nlink += get_nr_threads(p);
3703 put_task_struct(p);
3704 }
3705
3706 return 0;
3707}
3708
3709static const struct inode_operations proc_task_inode_operations = {
3710 .lookup = proc_task_lookup,
3711 .getattr = proc_task_getattr,
3712 .setattr = proc_setattr,
3713 .permission = proc_pid_permission,
3714};
3715
3716static const struct file_operations proc_task_operations = {
3717 .read = generic_read_dir,
3718 .readdir = proc_task_readdir,
3719 .llseek = default_llseek,
3720};